CHAPTER 4.
Method for Determining Skin Tone
Evenness and Skin Texture
Smoothness
57
4.1
Introduction
The study conducted by Ishida and Daibo (see Chapter 2) revealed that the perception of
skin translucency can be described in terms of three attributes: skin tone evenness, skin texture
smoothness, and the skin radiance. If these three attributes have a marked influence on the
perception of skin translucency in the judgment of consumers, we should be able to reproduce
the perception of skin translucency technically by measuring these three attributes. As a first
step in proving this concept, I aimed to develop objective measurements corresponding to each
of these three attributes. In this chapter, I will discuss the image-analysis method that I
developed to measure skin tone evenness and skin texture smoothness.
Skin texture originates from the three-dimensional microtopography of the skin, which
has “ridges” and “valleys” of 5 to 100 mm height and depth [59], known as micro-relief
patterns. Light incident on the skin is scattered by the curvature of the ridges of the skin,
whereas shadows are created along the valleys. This pattern is captured by the human eye as a
micro contrast in the luminance of the skin and is recognized as skin surface texture [60]. As
the ridges of the skin topography become higher and the valleys become deeper, in other
words, the skin surface texture becomes rougher, the contrast between light scattered by
ridges and the shadows in the valleys will become larger. The roughness of the skin texture is
therefore associated with the contrast in the micro-light scattering by the ridges and valleys.
Skin tone originates from the presence in the skin of chromophores such as melanin and
hemoglobin. These biological pigments absorb incident light in the skin layers and produce
the color of the skin [40]. The greater the concentration of chromophores that is present, the
darker or more chromatic the skin becomes. In particular, the presence of excessive melanin
in the skin causes a skin discoloration, known as a spot. A spot is typically generated after
exposure to ultraviolet (UV) radiation or appears gradually on the surface of the skin as a
result of aging [61]. A spot is recognized when excessive pigmentation occurs locally on the
58
skin area. If the hyperpigmentation occurs over the entire skin area, the skin simply becomes
darker. When the contrast of an area of localized pigmentation is strong, it is recognized as
spot. If the contrast is low, it is recognized as the skin tone dullness. Therefore, as in the case
of skin texture roughness, skin tone evenness is also associated with contrast, which depends
on the concentration of chromophores in the skin. The difference between skin texture
roughness and skin tone unevenness can be described in terms of the scale of this contrast.
As reviewed in Chapter 2, techniques exist for assessing the surface quality of skin by
means of Fourier analysis [50, 51, 52, 53]. Fourier analysis can convert contrast-based
variations into spatial frequencies and quantify these. Therefore, it is possible that Fourier
analysis could describe skin tone evenness and skin texture smoothness in a single common
description.
In the remainder of this chapter, I will discuss the application of Fourier analysis to the
measurement of skin tone evenness and skin texture roughness.
4.2
Objective
This remainder of this chapter will describe the development of an objective and
quantitative method for measuring skin tone evenness and skin texture smoothness by means of
Fourier analysis. To validate the method, the results of computations were compared with
results of visual evaluations made by human observers.
4.3
Image Analysis Program Development
The image-analysis program based on Fourier analysis was built within the MATLAB
computing environment. The algorithm involves five steps that will be introduced in this
59
section. The overall flow chart for the developed image analysis algorithm is shown in Figure
10, and the source code of the program is attached as Appendix B at the end of this thesis.
The analysis of the facial images is performed by means of the following five steps:
(1)
The ROI is cropped manually from the original images.
(2)
The cropped images are converted into eight-bit gray-scale images.
(3)
The gray-scale images are decomposed into frequency-domain components by using the
fft2 function in MATLAB.
(4)
A band-pass filter is applied to the converted images and a signal within a specific
frequency-domain range selected by the filter is extracted.
(5)
The power spectrum of the band-pass-filtered frequency regions is integrated. This step
permits quantification of features that correspond to selected frequency-domain regions.
4.3.1
Step 1: ROI Cropping
For the analysis of skin tone evenness and skin texture smoothness, images from REAL
3.0 were used, because these are captured under lighting conditions that mimic natural white
light and are the most suitable for evaluating skin tone and skin texture. Each REAL3.0 image
is originally 2016 × 3024 pixels and includes the entire face of the subject. The image is large
in size and contains features that are not related to skin tone and texture analysis, such as the
color charts for color correction placed near the chin of the subject (Figure 4) or facial features
that are not of interest (for example, the eyebrows, eyes, eyelashes, and mouth). To exclude
these superfluous features from the image analysis, the facial image was cropped around the
cheek area so that the cropped region included only the target features, giving an ROI of 512 ×
512 pixels (Figure 11) . The cheek area was chosen because it is the largest area of continuous
skin on the face and thus the easiest in which to find an area for the analysis that is not
disturbed by other features. Also, the cheek is the area of the face that is of most concern to
60
consumers, who look most frequently at the cheek area when they are evaluating the
appearance of facial skin. Therefore, analyzing the cheek region makes sense both from the
point of view of convenience of analysis and from that of relevance to the consumer.
4.3.2
Step 2: Gray Scale Conversion
The second step in the analysis was to read the image and to convert it into an eight-bit
gray scale. The rgb2gray function of MATLAB was used for this conversion. The image
cropped in Step 1 was a full-color image and therefore consisted of three individual color
channels (red, green, and blue; RGB). It might have been more informative to perform the
analysis on the full-color image, but the analysis of three channels would have increased the
volume of data to be analyzed and made the analysis more complicated. As the skin tone
evenness and texture roughness are detectable as variations in luminance contrast [62], I
chose to start with gray-scale images for the purpose of Fourier analysis.
4.3.3
Step 3: Performing the fast Fourier transform (FFT) Analysis
The cropped and gray-scaled image was ready for analysis by means of the Fourier
transform. The image was decomposed into frequency components by using the fft2 function
in the MATLAB Image Processing Toolbox. The fft2 function decomposes the original source
image, g(x,y), where x and y correspond to two perpendicular axes in real space, into its power
spectrum image, G(f,θ), where f is the frequency of the composite waves and q is the phase
angle, as shown in Figure 14, in accordance with the principles of the discrete Fourier
transform. In the power-spectrum image, a larger value of f corresponds to a higher frequency
and therefore a shorter wavelength. When the value of f is equal to 1, the composite wave has
the highest frequency that can be found in the image. As the original source image is a digital
image that consists of pixels, the minimum wave should have a two-pixel cycle. When the
61
value of f is equal to 0, this denotes the largest frequency and the longest wave. Such a wave
should have an infinite wavelength and is therefore a “constant” component. In this particular
facial skin analysis, a higher frequency corresponds to a more-micronized variation and
therefore the texture roughness should be described by higher-frequency components than the
skin tone unevenness, which should have lower-frequency components. The constant
component represented by the zero f value corresponds to the mean luminance of the images
and therefore describes the average skin tone.
4.3.4
Step 4: Band-Pass Filtering
The skin tone evenness and skin texture roughness are decomposed into regions of
different sizes by frequency in the power spectrum images. To extract these components from
the image and to separate them from each other, the power spectrum image has to be filtered
by means of a coaxial circular binary filter (the band-pass filter; Figure 13). The band-pass
filter produces a two-dimensional image that consists of binary data (0 or 1) and has the same
number of pixels as the original source image and the power spectrum image. To filter
specific frequency band from the power spectrum, corresponding pixels in each band-pass
filter image and power spectrum image are multiplied by each other. The pixels in the power
spectrum image that are multiplied by a zero value in the band-pass-filtered image lose their
pixel values, whereas pixels that are multiplied by pixels with a unit value in the band-pass
filter retain their original pixel value. In this way, the band-pass filter can extract a signal that
corresponds to a specific frequency band.
Skin tone unevenness should be filtered by a lower-frequency band whereas skin texture
roughness should be filtered by a higher-frequency band, because skin texture is a visibly
more-micronized variation in the luminance contrast of the image. The association of
particular frequency bands with these skin features will be discussed in Section 4.5.
62
4.3.5
Step 5: Quantification of the Power Spectrum
The filtered power spectrum image contains only bands for selected frequency
components. The power spectrum shows the intensity of these components and, therefore, the
integration of the filtered power spectrum can describe the total quantities of the
corresponding frequency components that are present in the original source image [53]. The
power spectrum was integrated by summing up all the data remaining in the filtered power
spectrum, as shown in Figure 14. The integrated value was then logarithmically transformed
for further analysis.
4.4
Program Validation by using Manipulated Images
The computing program was developed with the aim of extracting and quantifying skin
tone evenness and skin texture smoothness by means of the Fourier transform. This section
discusses the validation of the computer program from the following two points of view.
(1) The first is whether the program can extract features of the skin, such as spots (typical
skin tone unevenness) or pores (typical skin texture roughness) from the original source
image. This corresponds to a validation of the performance of the band-pass filtration
function.
(2) The second is whether the program can quantify the numbers of spots or pores. This
corresponds to a validation of the summation of the filtered power spectrum.
To conduct these validations, manipulated facial skin images that had different visibilities of
spots or pores were created by using Photoshop® CS3 (Adobe Systems Inc., San Jose, CA),
and the resulting images were analyzed by using the computer program.
63
4.4.1
Manipulated Image Creation
Two 512 × 512 pixel facial images cropped from REAL 3.0 images of Japanese females
who had markedly visible spots (Figure 15, S100) or pores (Figure 16, P100), respectively,
around the cheek area were used as the base images for the image manipulation. These images
were retouched by using Adobe® Photoshop® CS3 and the visible spots or pores were
completely removed from the image by using the healing brush tool. The resulting images are
the S0 and P0 images shown in Figure 15 and Figure 16, respectively. These S0 and P0
images were then overlaid on top of the S100 image and the P100 image, respectively, to
blend with each other. (the S0 and P0 images formed the upper layers and the S100 and P100
images formed the bottom layers). The opacities of the S0 and P0 images were varied from
100% to 0% in 10% increments. When the opacity was 100%, the S0 and P0 images were
totally opaque and completely masked the underlying S100 and P100 images. When the
opacity was 0%, the S0 and P0 images became completely transparent and S100 or P100
images underneath were fully visible. Between 0% and 100% transparency, the top and bottom
images were blended with each other and each resulting image gave a different visibility of
spots or pores. In this way, eleven images each with different and graded visibilities of spots
(Figure 15; S0 to S100) or pores (Figure 16; P0 to P100) were created.
4.4.2
Results
For the first purpose of the validation, the manipulated images were analyzed by using
the codes from Step 1 to 3 of the computer program. This resulted in the power spectrum
charts shown in Figure 17. To avoid the presence of dense lines, only the charts for the S0,
S100, P0, and P100 images are shown. As can be seen from these charts, changes in a specific
region of the power spectrum band were observed as a result of spot and pore manipulation.
This shows that the frequency band between 0.02 and 0.1 corresponds to spot manipulation
64
and that between 0.1 and 0.3 correspond to pore manipulation. Changes in spots (skin tone
unevenness) and pores (skin texture roughness) should therefore be separated by frequency;
lower spatial frequencies are responsible for skin tone unevenness, whereas higher spatial
frequencies are responsible for skin texture roughness. The frequency bands that
corresponded to spot manipulation (0.02 to 0.1) and pore manipulation (0.1 to 0.3) were then
used in the following band-pass filtering (Step 4 of the program), and the filters were
multiplied to the power-spectrum images of the skin. The resulting band-pass filtered
power-spectrum images were then inversed by ifft2 function in MATLAB to re-convert to
real-view images so that the result of band-pass filtering was confirmed visually. This
corresponds to conversion from image (F) to (G) in Figure 13. As the result, images shown in
Figure 18 were obtained. The band-pass filtered real-view images visually prove that the
features of the spots or pores can be isolated from the original image of the skin by using a
band-pass filter, and hence the computer program was validated in the first point of view.
Next, the power-spectrum in these selected frequency bands was quantified for the
second validation purpose. The results of this quantification are shown in Figure 19. The data
were plotted versus the opacity percentages of the S0 and P0 images. The S0 and P0 have
fewer spots and pores visible, respectively, whereas the S100 and P100 images show all the
original spots and pores. As can be seen in the plots, there was a good agreement between the
visibility of spots and pores (represented by the opacity of the layers) and the quantified
values obtained by fast Fourier analysis, and hence the computer program was also validated
from the second point of view.
These results confirmed that the developed computer program can filter features of skin
tone and texture and quantify them. Therefore, the next section describes my analysis of the
correlation between values from the Fourier transforms and the visual perceptions of skin tone
unevenness and skin texture roughness for natural images of skin (not manipulated ones) to
65
identify the frequency bands that are associated with these two skin features.
4.5
VPS of Skin Tone Evenness and Skin Texture
Smoothness
In the VPS, facial images of subjects were shown to respondents who assessed the
severities of skin tone unevenness and skin texture roughness in each presented image. The
same images were also analyzed by using the developed computer program, and scores from
the visual evaluation were compared with the computed values. From this comparison,
frequency bands corresponding to the perceptions of skin tone evenness and skin texture
smoothness were identified. By using the band-pass filters for the specified frequency regions,
the severity of the skin tone unevenness and skin texture roughness were quantified and then
the data were compared with the visual grading scores by means of a Pearson correlation to
determine how well image analysis can represent the visual perception of these skin features.
4.5.1
Subjects
The subjects participated in this study were 45 Japanese females (mean age 39.3 years; SD
7.9) from 20 to 49 years old.
4.5.2
Respondents
The respondents were 21 Japanese females (mean age 37.1 years; SD 4.4) who were
untrained to evaluate this kind of skin attributes and therefore the representative of
consumers.
4.5.3
Image Capturing
Images of the subjects were captured by using REAL3.0 from the left side of the face.
66
The procedure for the image capture was as described in Section 3.2.4. In this particular
analysis, the left side view images were used.
4.5.4
Study Procedure
The study procedure and the instrumental setup followed the absolute grading method
described in Section 3.2.3. The 45 images were presented sequentially in a random order on the
color-corrected LCD monitor (CLC202p; as described in Section 3.3.3). As the respondents
viewed the images, they were asked to give scores for skin tone unevenness and skin texture
roughness on a six-point grading scale (0: skin tone even or texture smooth, 6: skin tone very
uneven or texture very rough).
4.5.5
Searching for the Frequency Bands Responsible for Skin Tone
Unevenness and Skin Texture Roughness
The frequency band corresponding to the manipulated spots or pores was successfully
identified in Section 4.4. The manipulated spot or pore is an artificial change for computer
program validation, and the frequency band of the manipulated attributes does not necessarily
describe skin tone unevenness or skin texture roughness of consumer’s perception in the
natural situation. In order to measure unevenness of skin tone and roughness of skin texture
underlying the perception of skin translucency, similar analysis has to be carried out on
natural (non-manipulated) images. For this purpose, forty-five skin images that have various
degrees of skin tone unevenness and skin texture roughness were assessed by visual grading
by respondents in the VPS, and then the result was compared with frequency-domain
components of the forty-five images.
The results are plotted in Figure 20. The x-axis of the graph corresponds to the
wavelength (millimeter scale), which was converted from the frequency units produced by
67
MATLAB for easiness to understand how the scale of skin tone unevenness and skin texture
roughness looks like. The scale corresponds to the size of features that is to be filtered by the
corresponding band-pass filter. The y-axis of the graph indicates the Pearson correlation r
between each frequency component and the score from visual grading of the skin tone
evenness or skin texture roughness. The results showed that perception of skin tone
unevenness was highly correlated with components having a size of 2–5 mm, whereas the
perception of skin texture roughness was highly correlated with components having a size of 1–
2 mm. Therefore, band-pass filters that permitted filtration of these sizes of features were
designed and multiplied to the power-spectrum images to obtain a power-spectrum of skin
tone unevenness and the one of skin texture roughness. In the next section, by using these
band-pass filters, the skin tone unevenness and skin texture roughness are quantified.
4.5.6
Quantification of Skin Tone Unevenness and Skin Texture
Roughness and Correlation to the Visual Evaluation
The band-pass filter designed as described in Section 4.5.5 was applied to the
power-spectrum images processed from the skin images (refer to Figure 13), and the severity
of skin tone unevenness and skin texture roughness was quantified for each of the 45 images
as described in Section 4.3.5 (refer to Figure 14). The resulting indices of skin tone
unevenness and skin texture roughness were correlated with the scores from the visual
evaluation. The results are shown in Figures 21 and 22. The results show high degrees of
correlation between the visual grading scores as determined by visual grading and the
computed indices for skin tone unevenness and skin texture roughness as determined by the
Fourier analysis (r = 0.84 for both). This suggests that the method successfully filters these
skin features and quantifies their severity of skin tone unevenness and skin texture roughness.
68
4.6
Discussion
Attempts were made to measure skin tone unevenness and skin texture roughness by
means of the FFT, and a program for performing this analysis was developed on MATLAB
computing software. As seen in the preliminary applications done by Mifune [50, 51],
Hayashi [52], and Bargo [53] et al., the Fourier transform can decompose a two-dimensional
image of skin into consecutive special frequency components that correspond to the size of
features that appears on the skin. By applying this principle, I successfully extracted skin tone
unevenness and skin texture roughness from the frequency components by identifying the
frequency domains responsible for these skin features and by designing band-pass filters for
the these domains. The band-pass filter for the skin tone evenness consisted of
lower-frequency bands corresponding to the size of 2-5 mm wave in real scale whereas that
for the skin texture roughness consisted of higher frequencies corresponding to the size of 1-2
mm wave in real scale. Skin tone unevenness is associated with hyperpigmented spots or
localized skin tone discoloration described as skin dullness whereas skin texture roughness is
characterized by enlarged pores or micro-relief patterns on the surface of the skin. The results
for the frequency bands associated with skin tone unevenness and with skin texture evenness
correspond to this expectation: the band-pass filter used for the extraction of each attribute fell
in general range of typical spot or pore/texture size recognized by consumers, and therefore
the result makes sense from both of technical and consumer perception points of view.
The FFT program was also capable of quantifying skin tone unevenness and skin texture
roughness. To validate this function, artificially manipulated images of spots and pores were
created to provide tangible changes in these attributes in a manner that was suitable for
assessing the capability of the program to quantify these skin features. The quantification
process provided a reasonable approximation of the designed severity of spots and pores, as
69
shown in Figure 19, although quantification at low levels (between 0 and 10%) was poor in
both spot and pore detection. However, in a practical application to real images, the method
worked well, as shown in quantifying the difference of these skin features in natural images of
skin (not manipulated ones) very linearly with a high degree of correlation. The Pearson
correlation r = 0.84 would be a good match to the visually perceived score. Setaro and
Sparavigna has also evaluated irregularity of skin surface texture by means of the FFT
analysis and showed that their indices correlate with the age of the skin between ranges from
0.47 to 0.51 in terms of Pearson correlation r [63]. Comparing to their value, the r=0.84 of my
analyses is much higher and this indicates that the developed method for skin tone unevenness
and skin texture roughness works quite successfully.
4.7
Conclusion
A computing program based on a Fourier transform was developed and validated. The
program was useful for filtering skin tone unevenness and skin texture roughness and for
quantifying their severity. The quantification agreed with the visually perceived severity of
these attributes with a high degree of correlation. With these methods, two of the skin
attributes underlying the perception of skin translucency became measurable.
70
Figure 10: Fast Fourier Transform Flowchart
Overall flowchart for the program for the Fourier analysis of skin images. The program filters
the skin tone unevenness and skin texture roughness from images of skin captured by the
REAL3.0 system and then calculates the indices of severity of these skin features in the five
steps below.
71
Figure 11: Image Cropping Before Fourier Analysis (Step 1)
Facial images captured by using the REAL3.0 system (left) are cropped to give a 512 × 512
pixel image (right) as the first step of the Fourier-analysis.
Original REAL3.0 image
Cropped image for the FFT analysis
72
Figure 12: Production of Gray-Scale Image (Step 2)
The cropped image is converted into a gray-scale image before Fourier analysis. Even with
the loss of the color information, variations in the luminance of the skin, corresponding to
spots and pores, are visible in the gray-scale image.
73
Figure 13: Band-Pass Filtration of the Skin Image
This figure summarizes the procedure involved in band-pass filtration of the skin image. (A)
is the original source image to be analyzed. (B) is the power spectrum image of (A), obtained
by using the fft2 function of MATLAB. (C) is a summary of the power spectrum chart as a
function of f (see Figure 14 for more detail of this). (D) is an example of a band-pass filter; the
white area has a unit value (1) and the black area has a null value (0). This band-pass filter is
multiplied by image (B) to give image (F). (E) is the power spectrum of (F) and is equivalent
to the filtered power spectrum from (C). (G) is the inversed image back-produced from (F)
obtained by using ifft2 function of MATLAB. Image (G) shows the skin features remaining
after band-pass filtration. (a) denotes image processing by fft2 function in MATLAB that
produce a power-spectrum image from a real image, (b) denotes image processing by ifft2
(inverse of fft2) in MATLAB which back-produce a real image from power-spectrum image.
74
Figure 14: Quantification of the Power Spectrum
The original source image, g(x, y), [Left, please refer to image (A) in Figure 13] was
transformed into a power spectrum image, Ĝ(f, θ), [Middle, refer to image (B) in Figure 13]
by the Fourier analysis. The distance from the center (origin) of the power spectrum image
corresponded to the frequency of the components. To obtain the power spectrum chart of the
function of f [Right, please refer to images (C) and (E) in Figure 13], Ĝ(f,θ) was integrated by
q at each f. The power-spectrum function was then integrated between the frequency bands to
obtain the indices of the skin tone unevenness and skin texture roughness [Right, highlighted
in orange].
8 bit gray scale skin image
FFT frequency space (fx - fy plane)
FFT frequency spectrum
{
fy
2p
Gˆ ( f ) = ò G ( f , q )dq
g ( x, y) Þ G ( f , q )
0
f
g ( x, y )
FFT
}
log10 Gˆ ( f )
1.0
7.5
AUC(0.1, 0.4) / N
7
q
1.0
-1.0
8
fx
6.5
6
5.5
-1.0
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Spatial Frequency (f)
75
Figure 15: Artificially Manipulated Images (Spots)
The base image of skin (S100) was manipulated by using Photoshop, and the spots on the skin
were completely removed (S0 image). The S100 and S0 images were blended in Photoshop to
create images with various degrees of visibility of spots in terms of the Photoshop’s layer
transparency control in 10% increments.
S0
S10
S20
S30
S40
S50
S60
S70
S80
S90
S100 (base image)
76
Figure 16: Artificially Manipulated Images (Pores)
As with the spot images, the base image of skin (P100) was manipulated by using Photoshop
and the pores on the skin were completely removed (P0 image). The P100 and P0 images
were blended in Photoshop to create images with various degrees of visibility of pores in
terms of the Photoshop’s layer transparency control at 10% increments.
P0
P10
P20
P30
P40
P50
P60
P70
P80
P90
P100 (base image)
77
Figure 17: Changes in Frequency Components
The power spectrum function changes as the result of the manipulation of spots (top) and
pores (bottom) in the Photoshop images. The difference can be seen in the highlighted
frequency band for spots and pores, respectively, and these are the frequency bands that are
responsible for changing visibility as a result of manipulation of spots and pores.
7.5
7.3
7.1
FFT power
6.9
S0
S100
6.7
6.5
6.3
6.1
5.9
5.7
5.5
0
0.2
0.4
0.6
0.8
1
Frequency
7.5
7.3
7.1
FFT power
6.9
pore0
pore100
6.7
6.5
6.3
6.1
5.9
5.7
5.5
0
0.2
0.4
0.6
Frequency
78
0.8
1
Figure 18: Band-Pass-Filtered Images
By applying a band-pass filter to the original base image (Top), filtered images of the skin
texture (Bottom left) and skin tone unevenness (Bottom right) can be obtained. By comparing
these images, it was visually confirmed that the filtered images extract the desired features
from the original source image.
Original Base Image
Band-pass filtered image (texture)
Band-pass filtered image (spots)
79
Figure 19: Results of Image Analysis of Manipulated Images
The Photoshop-manipulated images for spots (top) and pores (bottom) were analyzed by
means of the Fourier transform and indices for the spots and pores were generated. The results
of the analysis agreed well with the visibility of spots and pores in the Photoshop-manipulated
images.
6.310
6.300
FFT - AUC/Sampling#
6.290
6.280
6.270
6.260
6.250
6.240
6.230
6.220
S0
S10
S20
S30
S40
S50
S60
S70
S80
S90
S100
Image Code
6.300
6.290
FFT - AUC/Sam plin g#
6.280
6.270
6.260
6.250
6.240
6.230
6.220
6.210
6.200
P0
P10
P20
P30
P40
P50
Image Code
80
P60
P70
P80
P90
P100
Figure 20: Comparison of Fourier Analysis and Visual Evaluation
Forty-five images were analyzed by using the Fourier-transform program and the power
spectrum value for each frequency was calculated. The power spectrum of each frequency
component was correlated with the visual evaluation scores for skin tone unevenness and skin
texture roughness, and the Pearson correlation was plotted against the spatial frequency. The
blue dots correspond to the correlations between each frequency and the visual evaluation of
skin tone unevenness determined in the VPS. The frequency band highlighted in blue showed
a high correlation with perceived skin tone unevenness. The frequency band highlighted in
red is similarly correlated with the perceived skin texture roughness.
0.9
Uneven skin color
0.8
Skin texture smoothness
Pearson Correlation
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
1
2
3
4
5
6
7
8
Spatial Frequency converted into corresponding wavelength [mm]
81
9
Figure 21: Correlation with Visual Evaluation (Skin Tone Unevenness)
The graph shows the comparison between skin tone unevenness as determined by visual
evaluation and that determined by Fourier analysis. The correlation coefficient was r = 0.84.
82
Figure 22: Correlation with Visual Evaluation (Skin Texture Roughness)
The graph shows a comparison between the skin texture roughness as determined by visual
evaluation and that determined by Fourier analysis. The correlation coefficient for these two
parameters was also r = 0.84.
83
CHAPTER 5.
Differences in Surface and
Subsurface Reflection
Characteristics of Facial Skin by
Age Group
84
Chapter 4 described how I developed objective measurements of skin tone evenness and
skin texture smoothness by means of the fast Fourier transform. In this chapter and the next, I
will discuss my attempts to develop methods for the objective measurement of skin radiance.
Eventually, I developed a technique for the measurement of skin radiance by the analysis of
surface and the subsurface components of reflection from the skin by using the SAMBA
system which will be discussed in Chapter 6. In this chapter, I first applied the SAMBA
technique on facial skin to confirm that the system is capable of evaluating optical properties
of skin. For this purpose, I chose to measure age-dependent characteristics of facial skin in
terms of the surface and subsurface reflectivities, and I discuss the association of these optical
parameters with other, more-conventional, measurements on skin.
5.1
Introduction
The appearance of facial skin changes with age in various ways, and in most cases it
deteriorates as more imperfections, such as hyperpigmented spots, wrinkles, enlarged pores, or
skin with a roughened texture, appear on the face. Changes in these skin attributes are more
obvious than skin radiance, and these have been attempted to be technically measured by
several objective methods, as reviewed in Chapter 1 and Chapter 2. I also developed an
objective method for the skin tone unevenness and skin texture roughness, which includes the
effect of hyperpigmented spots and enlarged pores, but what I achieved in the last Chapter
was a measurement of more extended skin qualities than single attribute of hyperpigmented
spot or enlarged pore.
In comparison with these skin features, optical attributes of the skin such as radiance, glow,
and shine are less tangible. The definitions of these consumer terminologies are not as clear as
85
those for the skin imperfections mentioned above, and their connection to physical parameters
is not fully established. Few objective measurements have been made of these optical attributes
in comparison with many other skin attributes [47, 64], and the visual grading method remains
the main tool for evaluating these attributes [65]. However, the optical appearance of the skin
can be quantified by determining the light-reflection profile of the skin. Gaining an
understanding of the optical-reflection characteristics by means of objective measurements is
an important first step in investigating less-tangible aspects of facial appearance such as
radiance, glow, and shine and the perception of the skin translucency.
The reflection of light from the skin is complicated by its multilayer structure [40]. The
stratum corneum, the outermost layer of the epidermis, is optically translucent and partially
reflects incident light while allowing most to penetrate to deeper layers of the skin. As the result,
the reflection from the skin is a mixture of specular surface reflection and diffuse subsurface
reflection [39]. It is therefore necessary to separate the surface and subsurface reflections from
the skin to achieve an understanding of the fundamental optical characteristics associated with
its appearance as previously tried by Masuda et al [39].
For the measurement of skin-color features, such as tone or hyperpigmented spots,
imaging systems equipped with two polarizing filters are widely used; the planes of
polarization of the filters are oriented perpendicular to one another, and one filter is located in
front of the source of illumination and the other is located front of the camera lens [36]. With
this polarized-light photography technique, the surface-reflection component from the skin,
known as specular reflection, is removed, and only the subsurface reflection component from
the skin is captured in the resulting image. If the planes of polarization of the two polarizing
filters are oriented parallel to one another, the plane of polarization of the removed
specular-reflection component is parallel to the plane of polarization of the incident light and
specular reflections are included in captured images. However, images captured with parallel
86
polarization also include subsurface-reflection components, so that the surface-reflection
component cannot be simply captured from images obtained by photography with polarized
light.
As a result of recent advances in electro-optical technology, polarization devices for which
the orientation of the plane of polarization can be changed by the application of an electrical
signal have been developed, and an imaging system equipped with such a device is
commercially available [56]. This equipment (SAMBA system) allows us to capture images for
parallel and perpendicular polarization states almost simultaneously; the surface reflection
components can then be extracted from the two resulting images by means of image processing,
because the two images for the different polarization states are captured for exactly the same
facial position. In this study, changes in the surface and subsurface reflection profiles of
Japanese women in three age groups (20–29, 30–39, and 40–49) were evaluated by using the
new imaging device and these profiles were compared with other skin parameters, such as the
hydration level, the melanin index, and the hemoglobin index.
5.2
Objective
In this chapter, the age-dependent changes of the surface and subsurface reflection profiles
of Japanese women in three age groups (20–29, 30–39, and 40–49) were evaluated to confirm
that the SAMBA system is capable of detecting changes in the optical quality of skin.
5.3
Materials and Methods
5.3.1
Subjects
The study involved 83 Japanese females in good general health aged 20 to 49 years. The
87
group aged 20–29 consisted of 15 females with an average age of 23.7 years (SD = 2.3). The
group aged 30–39 consisted of 30 females with an average age of 36.8 years (SD = 1.7). The
group aged 40–49 consisted of 38 females with an average age of 43.7 years (SD = 2.6).
5.3.2
Image Capturing Session
The image capturing session was performed by the standard procedure described in
Section 3.2.4. The SAMBA system was used to capture the facial images. In this study, the
images were captured from the front side of the face.
5.3.3
Image Processing
The principle and procedure of image processing by the SAMBA system is explained in
Section 3.3.2.
5.3.4
Skin-Condition Measurements
The level of skin hydration, the melanin index, and the hemoglobin index were also
measured for each subject. For the measurements of skin hydration, a Corneometer CM810
(Courage + Khazaka Electronic GmbH, Cologne) was used. The melanin and hemoglobin
indices were calculated from readings taken with a spectrophotometer (CMS-41XF3-3;
Murakami Color Research Laboratory, Tokyo) by using the formula derived by Takiwaki et al.
[66].
5.3.5
Data Normalization and Analysis of Age Dependency
The results for each individual subject were divided by the average for the 20–29 age
group for the purposes of normalization. The normalized data were statistically examined by
using one-way analysis of variance (ANOVA), followed by Fisher’s least significant
88
difference (LSD) as a post hoc test at 5% alpha.
5.4
Results
The surface and subsurface reflectivity characteristics of facial skin were evaluated for
two parameters: the mean and the SD. The mean reflectivity is an image histogram average
calculated from all the pixels in the rectangular area on the cheek. This parameter describes
the averaged “intensity” of the reflections, and a higher value corresponds to a higher intensity.
The SD represents the two-dimensional distribution of the reflection and describes its
“evenness” over all the pixels in the rectangular region. A higher SD corresponds to a more
uneven reflection profile.
5.4.1
Reflection Intensity
The surface reflection intensity gradually decreased with increasing age (Figure 23), but
the difference was not statistically significant. The subsurface reflection intensity decreased
significantly between the 20–29 group and the 30–39 and 40–49 groups (Figure 24).
5.4.2
Reflection Evenness
In the contrast to the reflection intensity, the evenness of the surface reflection showed a
significant change between the group aged 20–29 and the group aged 40–49 (Figure 25),
whereas the evenness of the subsurface reflection did not change with age (Figure 26).
5.4.3
Skin Hydration and the Melanin- and Hemoglobin-Indices
Changes in the skin hydration, the melanin index, and the hemoglobin index are plotted
in Figure 27 through Figure 29. A significant difference in the melanin index by age group
was observed, but no change was found in skin hydration or the hemoglobin index.
89
5.5
Discussion
The reflection of incident light from the surface of the skin occurs as a result of the
difference in refractive index between the air and the stratum corneum. It is known that,
typically, 4–8% of the incident light is reflected from the surface of skin [67]. The intensity of
the surface reflection is believed to be influenced by the refractive index of the stratum
corneum. The stratum corneum consists of layers of dead cells, and its main components are
keratin, lipids, and water. It is known that moisturized skin has a more radiant and brighter
appearance and Jiang also reported that dehydrated skin decreases its skin translucency [68].
Therefore I hypothesized that changes in the optical appearance of skin as a result of aging
might be manifested in the surface reflection intensity and, if such changes exist, that they
might arise from differences in skin hydration between age groups. However, in this study, I
found no significant age-dependent change in either the surface reflection intensity or the
level of hydration of the skin. Instead, I found a difference in the evenness of the surface
reflection. The results showed that light is reflected more unevenly by aged skin, whereas the
total reflection intensity remains unchanged. Landy [69] and Motoyoshi et al. [55] have
reported that visual perceptions of surface properties such as gloss or shine on vegetables or
works of sculpture can be explained in terms of simple statistics of the image histogram. Their
findings suggested that materials with occasional bright highlights surrounded by
predominantly dark areas appear to be glossier or shinier. Although their analyses were
performed on regular two-dimensional images with no differences in surface and subsurface
reflections, their interpretation can be applied to the optical appearance of skin. Aged skin has
a more uneven surface reflection and can thereby appear shinier and glossier than younger
skin.
90
Incident light that is not reflected at the boundary between the air and the stratum
corneum enters the skin. Some of this light is absorbed by chromophores such as melanin and
hemoglobin. The light that is not absorbed by chromophores reaches the dermis and is
reflected by collagen bundles. On the way back to the skin surface, some more light is
absorbed by chromophores and the rest emerges from the skin. The intensity and evenness of
subsurface reflections are therefore strongly influenced by the presence of chromophores. In
this study, melanin and hemoglobin indices were measured by means of spectrophotometer
readings and, as expected, the changes in the melanin index completely matched the changes
in the intensity of subsurface reflections. This shows that melanin in the skin absorbs light and
reduces the subsurface reflectivity, and that skin appears darker when there is more melanin
present.
It is interesting to note that differences by age group in the light-reflection profile occur
in the intensity of the subsurface reflection but not in its evenness, and that corresponding
differences occur in the evenness of surface reflection but not in its intensity. Subsurface
reflection makes up more than 90% of reflection from the skin and is the dominant
component of the reflected light. Subsurface reflection therefore determines the overall
brightness and tone of the skin. Also subsurface reflection occurs in a deeper layer of the skin
that is physically beneath the stratum corneum layers. This operates in a similar manner to the
backlight of an LCD monitor, so that the stratum corneum operates as a translucent diffusive
veil through which passes reflected light from the lower layers of the skin. This optically
complex layered structure creates unique qualities of skin appearance, such as translucency. In
contrast, surface reflection is a mirror-like specular reflection and if there is too much of this
component, the skin appears very unnatural. The combination of a high subsurface reflection
and an even surface reflection displayed by younger skin corresponds with the perception by
consumers that younger skin appears to be more radiant, as if glowing from the inside,
91
although its appearance is not shiny.
5.6
Conclusions
In this study, the light reflection from the skin was decomposed into surface reflection
and subsurface reflection, and the characteristics of these reflections were evaluated in
relation to the age group of the subject. Younger skin was found to show more subsurface
reflection and a more-even surface reflection, whereas aged skin loses its subsurface
reflectivity and reflects incident light more unevenly. These optical characteristics of the skin
could be related to the perception of shine, glow, or radiance.
The age-dependent changes of the subsurface and surface reflection from the skin were
successfully measured by using SAMBA system, and the reflective profile change made sense
from a biological perspective. These results support the view that the SAMBA system is
capable of measuring surface and subsurface reflection characteristics from the skin and can
be applied in the measurement of skin radiance. In the next chapter, a more-focused
application of this method to the perception of skin radiance will be discussed.
92
Figure 23: Age Dependency of Mean Surface Reflection
Changes with age group in the mean of the surface reflection, corresponding to the intensity of
reflection from the skin .
93
Figure 24: Age Dependency of the Mean Subsurface Reflection
Changes with age group in the mean subsurface reflection (the intensity of reflection from
inside the skin). Asterisks indicate differences that are significantly different (a = 0.05) from
the mean value for the 20–29 age group.
94
Figure 25: Age Dependency of the Standard Deviation of the Surface Reflection
Changes with age group in the SD of the surface reflection, which corresponds to the
unevenness of the reflective profile. The asterisk indicates a value that is significantly different
(a = 0.05) from that for the 20–29 age group.
95
Figure 26: Age Dependency of the Standard Deviation of the Subsurface Reflection
The histogram shows changes with age group in the SD of subsurface reflection (which
correspond to evenness of the subsurface reflection from the skin).
96
Figure 27: Age Dependency of Corneometer Readings
Changes in Corneometer reading by age group. No significant change with age group was
observed.
97
Figure 28: Age Dependency of the Melanin Index
Changes in melanin index by age group. The asterisks indicate values that are significantly
different (a = 0.05) from that for the 20–29 age group.
98
Figure 29: Age dependency of the Hemoglobin Index
Changes in hemoglobin index by age group. No significant change with age group was
observed.
99
CHAPTER 6.
Analysis of Perceived Skin
Radiance
100
A validation of the SAMBA method was described in Chapter 5. It was confirmed that
SAMBA system is capable of measuring and distinguishing the optical reflective properties of
the skin. This chapter is devoted to a discussion of optical properties of the skin that are
associated with the perception of skin radiance.
6.1
Introduction
Radiance, as an optical parameter defined by physics, can be measured in terms of the
amount of light that is emitted from a particular object, such as a light source. If we apply this
definition to skin radiance, then skin that reflects a larger proportion of incident light should
look more radiant. However, skin radiance as perceived by consumers is not necessarily defined
in this way. Skin radiance is a psychophysical parameter that involves quite complicated
surface and internal qualities of the skin [64], and it involves more than simply the quantity of
light that is reflected from the skin. People often mention that radiant skin appears to have an
internal glow. They distinguish radiant skin from shiny skin, from which they believe that the
majority of incident light is reflected from the surface of the skin. These views suggest that
people have the ability to perceive the depth of the reflection from the skin and to use this as a
basis for their perception of skin radiance. Does this ability really exist?
Even though people can tell whether incident light appears to be reflected from the surface
or from the inside of the skin, this does not provide direct proof that the human vision system is
capable of capturing reflected light from various depths within the skin. The skin consists of a
series of very thin layers, and human eyesight is insufficiently sensitive to recognize differences
in depth of the order of micrometers. It would more make sense to consider that human vision
perceives surface and subsurface reflections from the skin by capturing other features of the
reflected light from the skin. As discussed in Section 2.4, Motoyoshi et al. [55] have reported
101
that visual perceptions of surface qualities, such as the glossiness of a sculpture, can be
explained in terms of simple parameters of the image histogram, such as its skewness. They
created artificial images of a sculpture that appeared matte or glossy and they analyzed the
histograms of these images. They pointed out that a small area of brightness against a dark
background creates a perception of glossiness, and this can be described in terms of the
skewness parameter. Their study was based on a normal image consisting of both of surface and
subsurface reflections.
I thought that it would be quite interesting to expand Motoyoshi’s analysis to images of
facial skin that are split into its surface reflection and subsurface reflection components by
using the SAMBA system because, as described in section 2.4, the uniqueness of the skin
appearance originates in the complicated mixture of the surface and subsurface reflections. I
analyzed the surface and subsurface images of facial skin by means of their image histogram
parameters, and compared them with the visual evaluation by human observers for perceptions
of radiance. With these analyses, I eventually developed a mathematical regression model of
the human perceptions of skin radiance by using the image histogram parameters. This model
was capable of explaining the perception of skin radiance in technical measurement of
reflective characteristics of the skin. By combining with the metrics of the unevenness of skin
tone and the roughness of skin texture that were discussed in Chapter 4, a regression model
that can explain the perception of skin radiance by means of these measurements will be
developed in Chapter 7.
6.2
Objective
The remainder of this chapter is devoted to discuss an objective measurement for the
perception of skin radiance by using the SAMBA system. For this purpose, firstly it is
102
examined the correspondence between human perceptions of surface and subsurface
reflections from the skin and technical measurements of the surface and subsurface reflections
from the skin. Secondly then a mathematical regression model of the perceived skin radiance
in terms of the parameters of the technical surface and subsurface reflection image histogram
is developed.
6.3
Materials and Methods
6.3.1
Subjects
The subjects of this study were 45 healthy females aged 34 to 54 (mean age 45.9; SD 6.0).
6.3.2
Respondents
The respondents of this study were 30 females (mean age = 37.3; SD = 4.5). These were
naive respondents who were untrained in evaluating skin appearances before their enrollment in
the study and were therefore representative of female consumers.
6.3.3
Face-Imaging System
The SAMBA image capture and analysis system, as described in Section 3.3.2 , was used.
In this study, the facial images were captured from the left and right side of the face. The
image capture procedure was followed as described in Section 3.2.4.
6.3.4
VPS of Perceived Skin Radiance
The stimuli consisted of facial images of 45 subjects captured by using the SAMBA
system. For visual evaluation, full-color P-images were used. Two facial images of each subject
were captured with a four-week interval and prepared for evaluation. Over the four-week
103
interval, the subjects were instructed to use designated facial moisturizers in place of their
regular regimens. The perceptions of the 30 respondents of surface reflection, subsurface
reflection, and skin radiance of the two images presented side by side in a pairwise manner on a
color-corrected monitor (CLC202p, see Section 3.3.3) were recorded. I adopted a system of
comparative grading rather than an absolute grading because comparison of a pair of images
excludes variation of light reflectivity from the face that is oriented by the morphological
facial shape or skin color from the evaluation. Furthermore, we have no reference (guidance
scale) for judgment of the three attributes we examined, and assigning an absolute grading
score for these perceptions is more difficult psychologically than assigning comparative
differences between pairs of images.
The images of the 45 subjects appeared on the monitor in random order. In each visual
evaluation session, the respondent viewed the 45 pairs of images and replied to each of the three
questions below for each pair of images.
a) Which image do you think has more skin radiance?
b) Which image do you think shows reflections from the skin’s surface?
c) Which image do you think shows reflections from inside the skin?
To complete the evaluation for all three questions, three rounds of sessions were conducted
using the same image pairs. The respondents were instructed to evaluate the skin on the cheek,
because this is an area that is amenable to computer-based image analysis. The respondents
were asked to select the appropriate image in reply to the questions above and to give a score
corresponding to their degree of certainty according to the four ratings below:
4: Definitely
3: Moderately
2: Slightly
1: Maybe
104
Even if they thought there was no difference between the two images, they were forced to
choose one or other and to ascribe a rating of 1 in this case. In other cases, the choice of the
rating relied on their perceptual scale. The scores from the evaluation were automatically
recorded by the computerized system. A positive score was recorded when the later image
(captured at week 4) was chosen, whereas a negative score was recorded when the earlier image
(captured at week 0) was chosen. Therefore, if there was no difference between the two images
and the respondents were forced to choose ratings of 1, the respondents’ choices should be split
in half into the later or earlier image and the average from the 30 respondents would result in a
zero score, implying there was no difference between the two images. The data corresponding
to the answers to the three questions are denoted Y1, Y2, and Y3, respectively, as shown in
Figure 30.
6.3.5
Analysis of Image Histogram Parameters
The S-image (which included only surface reflections from the skin as shown in Figure 9
Surface Ref.) generated by image processing with SAMBA and the C-image (which included
only subsurface reflections from the skin as shown in Figure 9 C) were used in the analysis of
the image histogram parameters. Before the image analysis, these images were converted into
gray-scale images. A designated hexagonal region was cropped from each of the images so that
the hot spot on the cheek (the visibly brightest area that is the visible source of the perception of
radiance) was included in the ROI (Figure 31). The image histogram summarizes the
distribution of the signals in the pixels in the ROI. In addition to Motoyoshi’s published work
[55], Landy also mentions that human eyesight is sensitive to at least three properties of the
histogram (mean, variance, and skewness) [69], and I therefore chose these three properties as
parameters for my analysis. The correspondence between these image histogram parameters
and the appearance perceived by humans is modeled in Figure 32. The mean, the SD, and the
105
skewness were calculated for each of the surface reflection and subsurface reflection images, so
that we had six parameters in total. These were denoted X1 through X6, respectively, as shown
in Figure 30.
6.3.6
Partial Least Squares Regression Modeling
Partial least squares (PLS) regression analysis, as implement on JMP 8.0.2 software (SAS
Institute Inc., Cary, NC), was used to analyze the association between the respondents’
perceptions and the image histogram parameters. PLS regression analysis was chosen because
the independent variables were correlated with each other and problems of multicollinearity
would occur if a multiple regression was performed directly [70]. Among the benefits of PLS
regression analysis are that it can deal with multiple responses and that it can yield values of the
variable importance projection (VIP) that can serve an index for the statistical importance of the
independent variables in the final model. If a variable has a small coefficient and small VIP
value, it is a candidate for deletion from the model. Wold et al. [71] consider a value of less than
0.8 to be a small VIP value. In the first round of PLS regression analysis, all six technical
parameters were included in creating the model. VIP values of the technical parameters were
then checked and those with a value of less than 0.8 were excluded from the model. A second
round of PLS regression analysis was then performed for the variables with a VIP value of more
than 0.8. This process was repeated until all the technical variables that remained in the
regression model had VIP values of more than 0.8. When we reached this final model, we
examined the centralized coefficient to describe the contribution of each technical parameter to
the perceived optical skin features.
106
6.4
Results
In the first round of PLS regression analysis, three image histogram parameters had VIP
values in excess of 0.8 (Table 2). These three parameters were the mean of the subsurface
reflection (X1), the SD of the surface reflection (X5), and the skewness of the surface reflection
(X6). We then performed a second round of PLS regression analysis with these three technical
variables and we confirmed that the VIP parameters for all three parameters were more than 0.8
(Table 2). We therefore adopted the centralized coefficients obtained from the second-round
PLS regression analysis as a final model consisting of one subsurface-reflection parameter and
two surface-reflection parameters. The resulting coefficients are summarized in Table 3. The
centralized coefficient obtained by PLS regression analysis indicates the contribution of each
image histogram parameter to the visual perception attributes. The perceived subsurface
reflection (Y1) corresponds mainly to the mean of the subsurface reflection component, as the
coefficient is 0.744, whereas the contributions from the surface reflection components are 0.177
and 0.013, respectively. The perceived surface reflection (Y2) can be explained in terms of the
two surface reflection components, the coefficients for which are greater (0.512 and 0.420) than
that for the subsurface reflection component (0.273), and the contribution from these two
surface reflection components is almost equal. To explain the perception of skin radiance,
however, the mean of the subsurface reflection and one of the surface reflection components
(the SD) are needed, as the centralized coefficients for these two parameters (0.466 and 0.417)
are more weighted than that of the skewness of the surface reflection (0.291).
6.5
Discussion
The first objective of this study was to examine the correspondence between human
perceptions of surface and subsurface reflections from the skin and technical measurements of
107
the surface and subsurface reflections from the skin. The second objective of the study was to
develop a mathematical regression model of the perceived skin radiance in terms of the
parameters of the technical surface and subsurface reflection image histogram. To address these
objectives, I used PLS regression analysis. Respondents’ perceptions of subsurface reflection
can be mostly explained in terms of the mean from the image histogram for subsurface
reflections from the skin, whereas their perceptions of surface reflections can be well explained
in terms of two parameters of the image histogram (the SD and skewness) for surface reflection
from the skin. These results show that the respondents’ perception of the depth of light
reflection is actually associated with the characteristics of light reflection from different depths
within the skin, but what respondents use to recognize these attributes are the two-dimensional
descriptive statistics of the images rather than the actual depths. The mean of the image
histogram describes the average brightness of the measured region, and this represents the
intensity of reflected light captured in the image. The intensity of light from inside the skin is
determined by the density of chromophores such as melanin or hemoglobin in the skin, so that
the mean from the image histogram for subsurface reflections indicates the overall skin tone or
color. This suggests that people perceive fairer skin as showing more internal reflection. On the
other hand, the SD and skewness of the image histogram represent the characteristics of the
spatial distribution of the reflective light in the measured region. The association between the
skewness and the perception of the surface reflection is in line with the findings of Motoyoshi et
al. [14], who showed that a bright area surrounded by a dark region provides a perception that
the light is coming from the surface of the skin. Our research suggests that, in addition to the
skewness, the SD of the surface reflection is also involved in the recognition of skin surface
reflection.
Next, we created a model of skin radiance in terms of the parameters in the image
histogram corresponding to surface reflection and subsurface reflection. The coefficients of the
108
model showed that perceived skin radiance can be explained in terms of the subsurface
reflection and one of the surface reflection components from the skin. In the respondent’s mind,
skin radiance, a desired skin-beauty attribute, is associated more with subsurface reflections,
whereas surface reflections, which are often considered to be detrimental to skin beauty, are
connected more with skin shine. This may appear to contradict the PLS regression model, but
respondents actually describe the appearance of skin that does not show any surface reflection
as being matte and lacking radiance or shine. Our PLS regression model explains these
complicated perceptions of respondents. The perception of superficial reflection is simply
described by the surface reflection components (the SD and skewness) from the skin, whereas
the perception of skin radiance involves a well-balanced mixture of the characteristics of
tone/color (subsurface reflection) and half the surface reflection components (the SD), which
are also involved in the perception of superficial reflection. This explains why skin radiance has
remained an intangible quality for such a long time and is difficult to differentiate from skin
shine: the perception of skin radiance shares its visual quality partially with the perception of
skin shine but also involves elements of skin color and tone. If we simply increase surface
reflection by applying a reflective agent such as a moisturizer, the face may acquire a quality of
skin radiance but, at the same time, it can also acquire a quality of skin shine. Therefore, to
achieve an improvement in skin radiance, the addition of subsurface reflection is required and,
furthermore, careful control of the surface reflection components is also necessary.
6.6
Conclusion
Visual perceptions of surface reflection, subsurface reflection, and skin radiance are
associated with parameters of the image histogram. Respondents perceive subsurface
reflections on the basis of the mean statistics of the image and they perceive surface reflections
109
on the basis of the distributive statistics (SD and skewness) of the images. The perception of
skin radiance involves a mixture of both surface and subsurface reflection.
110
Figure 30: A flowchart for the Study.
P-images and C-images captured by using the SAMBA image-capture system were analyzed
with a computer-based image-analysis algorithm and by human observers in a visual
evaluation. The image analysis gave six objective variables (X1–X6) for subsurface and
surface reflections. The visual evaluation gave three subjective variables (Y1–Y3) for the
subsurface reflection, surface reflection, and skin radiance, respectively. PLS regression
analysis was performed to create a model that links the subjective variables with the objective
variables.
P-image
C-image
image processing (P - C)
S-image
C-image
Visual Evaluation
Image Analysis
Image Histogram Parameters
Subsurface components
Surface components
• mean (X1)
• mean (X4)
• s.d (X2)
• s.d (X5)
• skewness (X3)
• skewness (X6)
Perceptional Variables
• Subsurface reflection (Y1)
• Surface reflection (Y2)
• Radiance (Y3)
PLS regression analysis
111
Figure 31: SAMBA Images
The P-image (colored) was used for the visual evaluation of perceived surface reflection,
perceived subsurface reflection, and perceived skin radiance. The C-image (gray scale) and
the S-image (gray scale) were used in the image histogram analyses of the subsurface and
surface reflection components from the skin, respectively. The hexagonal area in the C-image
and S-image corresponds to the region of interest (ROI) used for the image histogram analysis.
The ROI was chosen to include the visibly brightest part and to avoid shadows on the cheek.
(a) P-image
(b) C-image
(c) S-image
112
Figure 32: Concept of the Image Histogram and the Image Histogram Parameters (Mean,
Standard Deviation, Skewness) Illustrated by Synthesized Images.
The image histogram is a chart that shows the distribution of pixel values in the image. The
images below are prepared to explain the concept of image histogram analyses by simulating
the appearances of skin. This chart shows how the image histogram parameters change with
changing appearance of the image.
113
Table 2: Variable Importance Projection (VIP) Values for the First and Second Rounds of
the PLS Regression Analysis
In the first round of PLS analysis, the VIP values for three parameters (X1, X5, and X6)
exceeded 0.8, the criterion for inclusion in the second round of analysis. These three
parameters all remained after a second round of analysis and were adopted as the final model.
Image histogram parameters
VIP value in Round 1
VIP value in Round 2
of the PLS analysis
of the PLS analysis
Mean of subsurface reflection (X1)
1.391
1.102
SD of subsurface reflection (X2)
0.781
n/a
Skewness of subsurface reflection (X3)
0.122
n/a
Mean of surface reflection (X4)
0.650
n/a
SD of surface reflection (X5)
1.497
1.080
Skewness of surface reflection (X6)
1.228
0.904
114
Table 3: Model Coefficients for Centralized Data
The size of the coefficient describes the contribution of each of the image histogram parameters
in the model against the corresponding perception. For example, the model equation for the
perceived radiance (Y1) can be described as Y1 = (0.466 * X1) + (0.417 * X5) + (0.291 * X6).
Reflection profile
Perceived subsurface
reflection (Y1)
Perceived surface
reflection (Y2)
Perceived radiance
(Y3)
Centralized coefficients for image histogram parameters
Mean of subsurface
SD of surface
Skewness of surface
reflection (X1)
reflection (X5)
reflection (X6)
0.714
0.177
0.013
0.273
0.512
0.420
0.466
0.417
0.291
115
CHAPTER 7.
Development of a Model of Skin
Translucency by Using Partial
Least Squares Regression
Analysis
116
7.1
Introduction
As discussed in Chapter 2, the perception of skin translucency is created by perceptions of
three more-tangible features of skin: skin tone evenness, skin texture smoothness, and the skin
radiance. Chapters 4, 5 and 6, describe how I developed new metrics for these three attributes
through Fourier analysis and measurements of the surface and subsurface reflections from the
skin.
In this chapter, I describe how I combined these objective measurements to build a
mathematical model of the perception of skin translucency. To do this, I obtained images of the
faces of 45 subjects and analyzed them by means of visual evaluation of skin translucency by
respondents and by objective measurements. By analyzing the resultant data through PLS
regression analysis (see Chapter 5), I obtained a mathematical model for the perception of
skin translucency. Having built the model, I validated it by using images of another 45
subjects; the skin translucency was again evaluated by visual evaluation and the model
equation was applied to the images to yield predicted values of the skin translucency index.
The two sets of results were compared by means of a Pearson analysis to determine whether
the predicted value and actual measured values showed a good correlation.
7.2
Objective
The objective of this research described in this chapter was to develop a mathematical
model of the perception of skin translucency and to validate the resulting model on a new
population of subjects.
117
7.3
Materials and Methods
7.3.1
Overall Test Design
Study 1 (Model Building)
The facial images of 45 subjects were captured by using REAL3.0 and SAMBA systems. The
REAL3.0 images were visually evaluated by 21 respondents in terms of the perception of skin
translucency in the VPS. The REAL3.0 images and SAMBA images were then analyzed for
skin tone unevenness, skin texture roughness, and the skin radiance as described in earlier
chapters. The association between the result of visual evaluation for perceived skin
translucency and the technical parameters were examined by means of PLS regression analysis
using JMP software to obtain a model equation that explains the perception of skin
translucency in terms of objective measurements of skin tone evenness, skin texture
roughness, and skin radiance (the mean of the subsurface reflection and the SD of the surface
reflection).
Study 2 (Model Validation)
Another 45 females were selected as subjects and their facial images were captured by using
REAL3.0 and SAMBA. The REAL3.0 images were evaluated by means of the VPS for
perceived translucency by ten untrained respondents who were different from those who
participated in Study 1. The REAL3.0 and SAMBA images were again analyzed by means of
the developed methods, and a technical index of skin translucency was calculated by using the
equation obtained in Study 1. This result was compared with the skin translucency as
determined by the ten respondents to check whether these two values correlated with each
other.
118
7.3.2
Subjects
The subjects for Study 1 were 45 females (mean age 39.3 years; SD 7.9). The subjects for
Study 2 were another group of 45 females (mean age 38.0 years; SD 8.4).
7.3.3
Respondents
The respondents for the VPS in Study 1 were 21 Japanese females (mean age 37.1 years;
SD = 4.4). The respondents to the VPS for Study 2 were ten Japanese females (ages
unrecorded). The respondents were naive, i.e. they were untrained in evaluating skin
appearance before their enrollment in the study, and were therefore representative of typical
female consumers.
7.3.4
VPS
The stimuli were facial images captured by using the REAL3.0 system. The images were
presented in a random order on the color-corrected LCD monitor (CLC202p; as described in
Section 3.3.3) and the observers were asked to provide a grading of perceived skin
translucency. In Study 1, the respondents were asked to look at the images and to give a score
corresponding to their perception of skin translucency on a four-point grading scale. A score of
4 corresponded to a strong agreement that the face in the image appeared to display skin
translucency whereas a score of 1 corresponded to a perception that the face in the image did
not show skin translucency. In the VPS in Study 2, the respondents were asked to score their
perceptions of skin translucency on a six-point grading scale, where a score of 6 corresponded
to the highest degree of skin translucency.
7.3.5
Objective Measurements
In both Study 1 and Study 2, the REAL3.0 images were also used in an analysis of skin
119
tone unevenness and skin texture roughness by means of the Fourier analysis developed in
Chapter 4. The subjects’ facial images captured by using SAMBA were used for the objective
analysis of skin radiance from the mean of the subsurface reflection and the SD of surface
reflection from the skin as discussed in Chapter 6.
7.4
Results
It was described in Chapter 6 how a PLS regression analysis is used to identify the
factors that are significantly associated with a dependent variable and it successfully
contributed to reveal the association of the perception of skin radiance with surface-reflection
and subsurface-reflection components from the skin. It also revealed difference between skin
radiance and shine in terms of the VIP values. In the current analysis performed in this chapter,
PLS regression modeling was again used to associate perceived skin translucency with the
results of the technical metrics that were measured on skin and I used the VIP value to assess
whether the metrics on skin can in reality be associated with perceived skin translucency. A
VIP value exceeding 0.8 was considered as a meaningful contributor to the model as
mentioned by Wold et al [71], and as used in Chapter 6. As the result, the resultant parameters
of the FFT corresponding to skin tone unevenness and skin texture smoothness, as well as the
mean of skin subsurface-reflection and the SD of skin surface-reflection, which correspond to
perceived skin radiance, all showed VIP values in excess of 0.8, proving that all these factors
contribute to the perception of skin translucency (Figure 33). Therefore all four parameters
were retained in the equation to create a final model for the perception of skin translucency. In
this way, we obtained a centralized coefficient that described the contribution of each optical
parameter to the overall perception of skin beauty, as well as a model equation that could be
used to back-calculate perceived skin translucency scores from the results of optical
120
measurements (Figure 34).
With these coefficients, the equation for perceived skin
translucency is described as follows:
SkinTransl ucency = -0.099243 ´ (TextureRou ghness) - 0.544972 ´ (ToneRoughn ess )
+ 0.3118025 ´ ( SurfaceMea n) + 0.3794965 ´ ( SurfaceSD )
To validate the equation, I conducted another round of image-shooting sessions (Study 2)
with REAL3.0 and SAMBA using 45 new female subjects. The FFT analysis and the image
histogram analyses of surface- and subsurface-reflections from the skin were performed on the
images of these 45 females, and the predicted skin translucency scores were calculated from the
results. In parallel, I conducted a visual-evaluation session to obtain perceived skin
translucency scores from ten naive respondents. These two results were compared and the
correlation between the predicted and perceived skin translucency scores was obtained. The
result was a Pearson correlation r = 0.72, which was statistically significant (Figure 35).
7.5
Discussion
The unpublished study by Ishida and Daibo demonstrated that perceived skin
translucency can be explained in terms of other skin perceptions, such as skin tone evenness,
skin texture roughness, and the skin radiance. Their analysis examined the relationship
between consumers’ perceptions of holistic skin appearance and their perceptions of
individual descriptive qualities of skin. I used this result as a basis for my studies to develop
objective measurement of unevenness of skin tone, roughness of skin texture, and the skin
radiance which was achieved by measuring the surface- and subsurface-reflections from the
skin. I developed objective measurements of these internalized skin qualities and evaluated
the results of these measurements in terms of the association between the results and
consumers’ perceptions of skin translucency. This showed that the measured metrics of skin
121
contribute significantly to consumers’ perceptions of skin translucency and that they can be
incorporated in a regression model of the holistic perception of skin beauty. It was my
hypothesis that, if the perception of skin translucency in the mind of the consumer is
associated with the three features of skin appearance (skin tone evenness, skin texture
roughness, and skin radiance), then that objective measurements of these three attributes
might explain the perception of skin translucency technically. PLS regression modeling and
evaluation from the VIP values proved that this hypothesis was true and that technical
measurements of these qualities can contribute to objective measurements of skin
translucency.
The VIP values obtained for the four parameters that I examined showed that the
contribution of skin tone unevenness was the highest (VIP = 1.42), followed by those of the
subsurface-reflection mean (VIP = 0.96), skin texture roughness (VIP = 0.94), and the SD of
surface-reflection (VIP = 0.86). An examination of the VIP values of these parameters shows
that skin tone unevenness makes the largest contribution whereas the contributions of the
other three are almost equal to one another. Skin tone unevenness is associated with the
presence of hyperpigmented spots or other local discolorations of the skin, which is oriented
in the uneven distribution of choromophores such as melanin or hemoglobin. Only less than
10% of light incidents to the skin is reflected at the surface or superficial layers of the skin,
and the rest penetrates into the deeper layers of the skin and interacts with the chromophores
[40]. Therefore, the influence of the subsurface-reflection should be a more marked change in
the skin compared with other features of skin. In the market for skincare products, skin
translucency is often associated with skin whitening, which controls the production of
melanin to give a skin tone that appears brighter and more even. Matts et al. say the color
homogeneity influence on the perception of age and youth, and the more homogeneous facial
skin color is perceived more attractive [72]. From biological point of view, Iwai et al. report
122
that the protein carbonylation in the stratum corneum, which is occurred over skin aging,
induces the decrease of skin transparency due to reduction of subsurface reflection from the
skin [73]. Therefore, the identification of skin tone unevenness as making the biggest
contribution to the perception of skin translucency makes sense and is understandable from
the both the technical point of view and from the point of view of consumers’ habits.
The model coefficients shown in Figure 34 are the actual coefficient values in the
mathematical model. The coefficients for skin tone unevenness and skin texture roughness are
negative, whereas those for the optical parameters of the skin are positive. This shows that for
skin to appear more translucent, the skin tone and skin texture must be more even (and
therefore the values of these measurement will be smaller) and the optical reflection
characteristics (mean of the subsurface reflection, SD of the surface reflection) have to be
larger. These tendencies are in line with everyday experience, confirming that the
mathematical model was properly created.
In the model validation (Study 2), the mathematical model was double checked with a
brand new population to confirm that the developed equation performs with not only the skins
of those subjects who were used for the model building, but also with those of a wider
population. As a result, I obtained a significant correlation (r = 0.69) between the skin
translucency predicted from the mathematical model and the actual visual evaluation score
from ten naive respondents. This suggests that the mathematical model is suitable for
evaluating the skin translucency of a wide range of group of consumers and can therefore be
applied to skin measurement as well as to the evaluation of the performance of skincare
treatments or newly developed skincare technologies. As a reference, Nakajima attempted to
describe skin translucency by only measuring the optical characteristics of the skin and
compared it with visual evaluation score. Her study showed a Pearson correlation r=0.490 to
the diffuse reflection component and the one r=0.282 to the specular reflection component
123
[49]. This suggests that by using the metrics of three skin attributes, skin tone evenness, skin
texture roughness and the skin radiance, better correlation to visual perception can be
obtained, and the modeling by means of these objective measurements successfully describes
the perception of skin translucency.
7.6
Conclusion
Facial skin beauty, described as translucency by consumers, is a holistic quality that
involves multiple attributes of the skin, such as skin tone evenness, skin texture smoothness,
and skin radiance. The perception of holistic beauty can be explained in terms of objective
measurements of these skin attributes, and I have developed a mathematical model that can
predict perceived skin translucency. By utilizing this method, we can now assess skin
translucency objectively and thereby contribute to the welfare of consumers by developing
superior products that make their skin appear more beautiful.
124
Figure 33: Importance of Skin Parameters for the Perception of Skin Translucency
The VIP values for the four technical metrics developed in Chapter 4 and 6 were examined in
the PLS regression analysis to create a technical model of the perception of skin translucency.
The result shows that VIP values of these parameters exceed 0.8, meaning that all these
factors contribute to the perception of skin translucency.
125
Figure 34: The Model Coefficients of the Skin Measurement Parameters for the
Perception of Skin Translucency
The actual coefficient values in the mathematical model. The coefficients for skin tone
unevenness and skin texture roughness are negative, whereas those for the optical parameters
of the skin are positive. This shows that for skin to appear more translucent, the skin tone and
skin texture must be more even (and therefore the values of these measurement will be
smaller) and the optical reflection characteristics (mean of the subsurface reflection, SD of the
surface reflection) have to be larger. These tendencies are in line with everyday experience,
confirming that the mathematical model was properly created.
Skin parameter
Coefficient
Roughness of skin texture (FFT)
-0.099243
Unevenness of skin tone (FFT)
-0.544972
Radiance (Mean of subsurface reflection)
0.3118025
Radiance (SD of surface reflection)
0.3794965
126
Figure 35: Model Validation by Means of the Pearson Correlation
In the model validation (Study 2), the mathematical model was double checked with a brand
new population. As a result, a significant correlation (r = 0.69) between the skin translucency
predicted from the mathematical model and the actual visual evaluation score from ten naive
respondents was obtained. This suggests that the mathematical model is suitable for
evaluating the skin translucency of a wide range of group of consumers.
127
CHAPTER 8.
Conclusion
128
By means of a series of studies, I successfully developed an objective evaluation tool that
can measure perceptions of intangible skin beauty (or skin translucency) by means of
objective measurements on facial skin used in conjunction with a mathematical PLS
regression model. I showed that perceived skin translucency can be described in terms of skin
tone unevenness, skin texture roughness, and the skin radiance, both in psychophysical
studies with consumers and in technical measurements. Objective methods for quantifying
these three skin attributes were successfully developed, and by measuring these three skin
qualities and incorporating the results in a PLS regression modeling equation, it was possible
to predict the translucency of facial skin. The developed model was confirmed to show a high
degree of correlation with observers’ visual perceptions of skin translucency.
To the best of my knowledge, this is the first method that has been successfully
developed for evaluating attributes of intangible skin beauty analytically and that can yield a
score that correlates high with the human perception of skin translucency. By using a tool that
is capable of evaluating skin translucency objectively, we can develop new and superior
technologies that can improve the beauty of skin. The evaluation method that I have
developed will be used in the evaluation of the performance of skincare products and in
assessing changes in skin quality arising from the application of these products.
This study was performed on Japanese subjects, but it is highly likely that its results will
also apply to the skins of other females of northeast Asian origin. The next logical step in the
research program will be to expand this method to persons from other regions and ethnic
groups.
129
References
130
[1] B. Fink, K. Grammer and P. J. Matts, "Visible skin color distribution plays a role in the
perception of age, attractiveness, and health in female faces," Evolution and Human
Behavior, vol. 27, p. 433–442, 2006.
[2] U. Yoshikazu, M. Behne, D. Quiec, P. M. Elias and W. M. Holleran, "Vitamin C stimulates
sphingolipid production and markers of barrier function in submerged human keratinocyte
cultures," Journal of Investigative Dermatology, vol. 117, no. 5, pp. 1307-1313, 2001.
[3] Y. Yazawa and R. Utsugi, "Reduction effect of pores by pre vitamin C and glycolic acid,"
Fragrance Journal, vol. 2, pp. 54-58, 2002.
[4] A. K. Chakraborty, Y. Funasaka, M. Komoto and M. Ichihashi, "Effect of arbutin on
melanogenic proteins in human melanocytes," Pigment Cell Research, vol. 11, no. 4, pp.
206-212, 1998.
[5] A. Kawada, N. Konishi, T. Momma, N. Oiso and S. Kawara, "Evaluation of anti-wrinkle
effects of a novel cosmetic containing retinol using the guideline of the Japan Cosmetic
Industry Association," The Journal of Dermatology, vol. 36, no. 11, pp. 583-586, 2009.
[6] T. Hakozaki, L. Minwalla, J. Zhuang, M. Chhoa, A. Matsubara, K. Miyamoto, A.
Greatens, G. G. Hillebrang, D. L. Bisset and R. E. Boissy, "The effect of niacinamide on
reducing cutaneous pigmentation and suppression of melanosome transfer," British
Journal of Dermatology, vol. 147, no. 1, pp. 20-31, 2002.
[7] D. L. Bisset, K. Miyamoto, P. Sun and C. A. Berge, "Topical niacinamide reduces
yellowing, wrinkling, red blotchiness, and hyperpigmented spots in aging facial ski,"
International Journal of Cosmetic Science, vol. 26, no. 5, pp. 231-238, 2004.
[8] B. Cravello and A. Ferri, "Relationships between skin properties and environmental
parameters," Skin Research and Technology, vol. 14, pp. 180-186, 2008.
[9] D. Black, A. Del Pozo, J. M. Lagarde and Y. Gall, "Seasonal variability in the biophysical
properties of stratum corneum from different anatomical sites," Skin Research and
Technology, vol. 6, no. 2, pp. 70-76, 2000.
[10] S. Nouveau-Richard, W. Zhu, Y. H. Li, Y. Z. Zhang, F. Z. Yang, Z. L. Yang, S. Lian, B. Y.
Qian, Y. P. Ran, C. Bouillon, H. D. Chen and O. de Lacharrière, "Oily skin: specific
features in Chinese women," Skin Research and Technology, vol. 13, pp. 43-48, 2007.
[11] S. Verdier-Sévrain, F. Bonté and B. Gilchrest, "Biology of estrogens in skin: implications
for skin aging," Experimental Dermatology, vol. 15, no. 2, pp. 83-94, 2006.
[12] M. J. Thornton, "The biological actions of estrogens on skin," Experimental Dermatology,
vol. 11, no. 6, p. 487–502, 2002.
[13] J. Hosoi, "Stress and the skin," International Journal of Cosmetic Science, vol. 28, no. 4,
131
pp. 243-246, 2006.
[14] M. Takahashi, "Non-invasive techniques to measure physiological properties of the skin
and their application to efficacy test of skin care products," Journal of Society of Cosmetic
Chemists of Japan, vol. 34, pp. 5-24, 2000.
[15] M. Takahashi, "Recent progress of skin bioengineering and its application to
substantiation of efficacy in cosmetics," Fragrance Journal, vol. 1, pp. 11-21, 1999.
[16] G. Imokawa, "The roles of stratum corneum components in the water-holding function of
the stratum corneum," Fragrance Journal, vol. 17, pp. 27-39, 2000.
[17] F. Li, E. Conroy, M. Visscher and R. Wickett, "The ablility of electrical measurements to
predict skin moisturization. I. Effects of NaCl and glycerin on short-term measurements,"
Journal of Cosmetic Science, vol. 52, p. 13–22, 2001.
[18] B. J. M. Morrison, "Comparison of instrumental measurements of skin hydration," Journal
of Toxicology, vol. 15, no. 4, pp. 305-314, 1996.
[19] K.-I. O‘goshi and J. Serup, "Skin conductance: Validation of Skicon-200EX® compared
to the original model, Skicon-100®," Skin Research and Technology, vol. 13, p. 13–18,
2007.
[20] H. Arimoto, M. Egawa and Y. Yamada, "Depth profile of diffuse reflectance near-infrared
spectroscopy for measurement of water content in skin," Skin Research and Technology,
vol. 11, pp. 27-35, 2005.
[21] H. Arimoto, "Estimation of water content distribution in the skin using dualband
polarization imaging," Skin Research and Technology, vol. 13, pp. 49-54, 2007.
[22] P. J. Caspers, G. W. Lucassen, E. A. Carter, H. A. Bruining and G. J. Puppels, "In vivo
confocal raman microspectroscopy of the skin: Noninvasive determination of molecular
concentration profiles," Joournal of Investigative dermatology, vol. 116, no. 3, pp.
434-442, 2001.
[23] S. Sakai, S. Sasai, Y. Endo, K. Matsue, H. Tagami and S. Inoue, "Characterization of the
physical properties of the stratum corneum by a new tactile sensor," Skin Research and
Technology, vol. 6, pp. 128-134, 2000.
[24] G. B. Jemec and J. Serup, "Epidermal hydration and skin mechanics," Acta Derm Venereol
(Stockh), vol. 70, pp. 245-247, 1990.
[25] H. S. Ryu, Y. H. Joo, S. O. Kim, K. C. Park and S. W. Youn, "Influence of age and regional
differences on skin elasticity as measured by the Cutometer," Skin Research and
Technology, vol. 14, p. 354–358, 2008.
[26] H. Dobrev, "Application of Cutometer area parameters for the study of human skin
132
fatigue," Skin Research and Technology, vol. 11, no. 2, p. 120–122, 2005.
[27] S. Ahn, S. Kim, H. Lee, S. Moon and I. Chang, "Correlation between a Cutometer® and
quantitative evaluation using Moire topography in age-related skin elasticity," Skin
Research and Technology, vol. 13, no. 3, pp. 280-284, 2007.
[28] S. Sakai, S. Sasasi, Y. Endo, K. Matsue, H. Tagami , S. Inoue, “Characterization of the
physical properties of the stratum corneum by a new tactile sensor,” Skin Research and
Technology, 第 巻 6, 第 3, pp. 128-134, 2001.
[29] K. D. Paepe, E. Houben, R. Adam, F. Wiesemann and V. Rogiers, "Validation of the
VapoMeter, a closed unventilated chamber system to assess transepidermal water loss vs.
the open chamber Tewameter®," Skin Research and Technology, vol. 11, no. 1, pp. 61-69,
2005.
[30] H. Takiwaki, Y. Miyaoka and S. Arase, "Graphic analysis of the relationship between skin
colour change and variations in the amounts of melanin and haemoglobin," Skin Research
and Technology, vol. 8, no. 2, p. 78–83, 2002.
[31] I. S. Yun, W. J. Lee, D. K. Rah, Y. O. Kim and B.-y. Y. Park, "Skin color analysis using a
spectrophotometer in Asians," Skin Research and Technology, vol. 16, no. 3, pp. 311-315,
2010.
[32] M. L. Abella, "Evaluation of anti-wrinkle efficacy of adenosine-containing products using
the FOITS technique," International Journal of Cosmetic Science, vol. 28, no. 6, pp.
447-451, 2006.
[33] S. Jaspers, H. Hopermann, G. Sauermann, U. Hoppe, R. Lunderstädt and J. Ennen, "Rapid
in vivo measurement of the topography of human skin by active image triangulation using
a digital micromirror device," Skin Research and Technology, vol. 5, no. 3, pp. 195-207,
1999.
[34] D. Nita, J. Mignot, M. Chuard and M. Sofa, "3-D profilometer using a CCD linear image
sensor: application to skin surface topography measurement," Skin Research and
Technology, vol. 4, pp. 121-129, 1998.
[35] M. Takahashi, "Image analysis of skin surfce contour," Acta Derm Venereol (Stockh), vol.
185, pp. 9-14, 1994.
[36] K. Miyamoto, H. Takiwaki, G. G. Hillebrand and S. Arase, "Development of a digital
imaging system for objective measurement of hyperpigmented spots on the face," Skin
Research and Technology, vol. 8, p. 227–235, 2002.
[37] K. Miyamoto, H. Takiwaki, G. G. Hillebrand and S. Arase, "ation of a high-resolution
digital imaging system for the objective and quantitative assessment of hyperpigmented
spots on the face," Skin Research and Technology, vol. 8, no. 2, pp. 73-77, 2002.
[38] F. A. Nathan, K. Jensen, H. Knaggs and S. Ferguson, "The clinical utility of a hand-held
computerized optical imaging system at assessing skin condition," Journal of Cosmetic
133
Dermatology, vol. 9, pp. 103-107, 2010.
[39] Y. Masuda, N. Kunizawa and M. Takahashi, "Methodology for evaluation of skin
transparency and the efficacy of an essence that can improve skin transparency," Journal
of Society Cosmetic Chemists Japan, vol. 39, no. 3, pp. 209-213, 2005.
[40] A. Krishnaswamy and G. V. Baranoski, "A biophysically-based spectral model of light
interaction with human skin," Eurographics, vol. 23, no. 3, pp. 331-340, 2004.
[41] T. Kawai and Y. Kurioka, "Volumetric visualization of human skin," IEEJ Transactions on
Electronics, Information and Systems, vol. 123, no. 11, pp. 1936-1943, 2003.
[42] J. M. Waller and H. I. Maibach, "Age and skin structure and function, a quantitative
approach (I): blood flow, pH, thickness, and ultrasound echogenicity," Skin Research and
Technology, vol. 11, no. 4, pp. 221-235, 2005.
[43] K. Kumasaka, "Functional analysis of the stratum corneum," 香粧会誌, vol. 15, no. 4, pp.
254-260, 1991.
[44] M. Saito, "Preference for fair skin in Asia with its psychological background," Journal of
the Color Science Association of Japan, vol. 29, no. 1, p. 46–53, 2005.
[45] N. Kashibuchi, S. Kimura and M. Suzuki, "Visible light and cosmetics (I) The effect of
cream and lotion on the transparency of stratum corneum," Journal of Japanese Cosmetic
Science Society, vol. 1, p. 9, 1985.
[46] M. Ooe and H. Shiroshita, "The relationship between transparent skin and optical
properties of stratum corneum," Fragrance Journal, vol. 4, pp. 38-44, 2002.
[47] Z.-x. Jiang and P. D. Kaplan, "Point-spread imaging for measurement of skin translucency
and scattering," Skin Research and Technology, vol. 14, p. 293–297, 2008.
[48] T. Seiya, M. Nomura, S. Hayashi and T. Hasegawa, "The structure of consciousness and
the physical characteristics of the skin regarding skin transparency," Journal of the Society
of Cosmetic Chemists Japan, vol. 38, no. 2, p. 115–124, 2004.
[49] N. Nakajima, "Optical reflection on skin surface correlated with visual factors such as
transparency and glossiness," Fragrance Journal, vol. 1, pp. 63-65, 2007.
[50] H. Mifune, "The feature extraction of human skin for computer graphics," In abstract of:
the 36th meeting of Informatino Processing Society of Japan, pp. 2081-2082, 1988.
[51] H. Mifune, K.-i. Okada, T. Yokoyama and M. Kitagawa, "The feature extraction of human
skin for computer graphics (2) - the synthesis of skin texture -," In abstract of: the 38th
meeting of Information Processing Society of Japan, vol. 38, pp. 717-718, 1988.
[52] T. Hayashi, K. Mimura and Y. Nishijima, "Changes in Surface configulation of skin caused
by aging and application of cosmetics.," Journal of the Society of Cosmetic Chemists
134
Japan, vol. 23, p. 43–54, 1989.
[53] P. R. Bargo and N. Kollias, "Measurement of skin texture through polarization imaging,"
British Journal of Dermatology, vol. 162, pp. 724-731, 2010.
[54] Y. Shibue, O. Osanai, E. Takeuchi, H. Iwai and Y. Minematsu, "The new evaluation of skin
color and skin tone on the facial skin," Journal of Society of Cosmetic Chemists of Japan,
vol. 26, no. 2, pp. 120-130, 1992.
[55] I. Motoyoshi, S. Nishida, L. Sharan and E. H. Adelson, "Image statistics and the
perception of surface qualities," Nature, vol. 447, p. 206–209, 2007.
[56] P. Clemenceau, S. Breugnot and B. Pouet, "In vivo quantitative evaluation og gloss,"
Cosmetic & Toiletries magazine, vol. 119, no. 10, pp. 71-78, 2004.
[57] N. Kollian, "Polarized light photography of human skin," in Bioengineering of the skin:
skin surface imaging and analysis, Boca Raton, Florida, USA, CRC Press, 1997, pp.
95-104.
[58] E. J. Bae, S. H. Seo, Y. C. Kye and H. H. Ahn, "A quantitative assessment of the human
skin surface using polarized light digital photography and its dermatologic significance,"
Skin Research and Technology, vol. 16, no. 3, pp. 270-274, 2010.
[59] V. Schreiner, G. Sauermann and U. Hoppe, "Chaqracterization of the skin surface by
ISO-parameters of microtopography," in Bioenginnering of the skin: Skin surface imaging
and analysis, Boca Raton, Florida, USA, CRC Press, 1997, pp. 129-143.
[60] K. Minami, M. Ito, N. Ojima, T. Minami and M. Hattori, "Optical determination of skin
texture," IFSCC magazine, vol. 3, no. 2, pp. 5-9, 2000.
[61] F. Minoru, N. Midori, A. Munakata, K. Nakajima and S. Ohta, "Effects of biological and
physical factors on ultraviolot erythemal and pigmentary response," Journal of Societry
Cosmetic Chemists Japan, vol. 13, no. 2, pp. 20-28, 1979.
[62] A. Sparavigna and R. Marazzato, "An image-processing analysis of skin textures," Skin
Research and Technology, vol. 16, pp. 161-167, 2010.
[63] M. Setaro and A. Sparavigna, "Irregularity skin index (ISI): a tool to evaluate skin surface
texture," Skin Research and Technology, vol. 7, pp. 159-163, 2001.
[64] A. Petitjean, J. M. Sainthillier, S. Mac-mary, P. Muret, B. Closs, T. Gharbi and P. Humbert,
"Skin radiance: How to quantify? Validation of an optical method," Skin Research and
Technology, vol. 13, p. 2–8, 2007.
[65] C. Musnier, P. Piquemal, Beau P and J. Pitter, "Visual evaluation in vivo of 'complexion
radiance' using the C.L.B.T sensory methodology," Skin Research and Technology, vol. 10,
pp. 50-56, 2004.
135
[66] H. Takiwaki , L. Overgaard and J. Serup, "Comparison of narrow-band reflectance
spectrophotometric and tristimulus colorimetric measurement of skin color," Skin
Pharmacol., vol. 7, p. 217–225, 1994.
[67] N. Kollias, "Polarized light photography of human skin," in Bioengineering of the skin:
skin surface imaging and analysis, Boca Raton, Florida, USA, CRC Press, 1997, pp.
95-104.
[68] Z.-X. Jiang and J. DeLaCruz, "Appearance benefits of skin moisturization," Skin Research
and Technology, vol. 17, pp. 51-55, 2011.
[69] M. S. Landy, "A gloss on surface properties," Nature, vol. 447, p. 158–159, 2007.
[70] I. E. Frank and J. H. Friedman, "A statistical view of some chemometrics regression tools,"
Technometrics, vol. 35, p. 109–135, 1993.
[71] S. Wold, L. Eriksson and S. Clementi, "Statistical validation of QSAR results," in
Chemometric Methods in Molecular Design, Weinheim, Germany, VCH, 1995, p. 309–
338.
[72] P. J. Matts, B. Fink, K. Grammer and M. Burquest, "Color homogeneity and visual
perception of age, health, and attractiveness of female facial skin," Journal of American
Academy of Dermatology, 2007.
[73] I. Iwai, T. Kuwahara and T. Hirao, "Decrease in the skin transparency induced by protein
carbonylation in the stratum corneum," Journal of Society of Cosmetic Chemists of Japan,
vol. 42, no. 1, pp. 16-21, 2008.
[74] A. Petitijean, J. Sainthillier, S. Mac-Mary, P. Muret, P. Closs, T. Gharbi and P. Humbert,
"Skin rasiance: how to quantify? Validation of an optical method," Skin Research and
Technology, vol. 13, pp. 2-8, 2007.
136
APPENDIX A
List of Abbreviations Used in the Text
ANOVA
Analysis of Variance
CIE
International Communications on Illumination
FT
Fourier Transform
FFT
Fast Fourier Transform
LCD
Liquid Crystal Display
LSD
Least Significant Difference
PLS
Partial Least Squares
ROI
Region of Interest
SD
Standard Deviation
VIP
Variable Importance Projection
VPS
Visual Perception Study
137
APPENDIX B
Source Code of MATLAB for the FFT analysis
functionpgfft
%ppm=p
ixelp
ermil
imeterofim
age
%Standa
rdFFTprogr
am
%ファイル名取得が GUI
%バッチ処理機能
%周波数空間で積分
%保存ファイルをセルで管理
%結果を対数で表示
%画像読み込み
% X:読み込み元画像
% Mothe
rDir:カレントディレクトリ
[X,Moth
erDir,finde
x]=uig
etfile
({'*.b
mp;*.j
pg;*.
tiff'},'
ファイル
を開く','Mult
iSelec
t','o
n');
path(pat
h,Mot
herDir
);
ppm=12.6
;
iffinde
x==0
;
disp('
ファイルが選択されていません')
;
else
%le
ngth(X
)=処理するファイルの数
SumR
esult=cell
(207,len
gth(X)
+1,4); % サマリーデータを
格納するセル (
202+5Xファイル数+1X4次元 (
RG
B+Gray))
SumR
esult{
1,1,1}='
File_W
avelen
gth[mm
]';
SumR
esult{
1,1,2}='
File_W
avelen
gth[mm
]';
SumR
esult{
1,1,3}='
File_W
avelen
gth[mm
]';
SumR
esult{
1,1,4}='
File_W
avelen
gth[mm
]';
form=1:4
SumR
esult{
203,1,
m}='
LightS
pot_CL
12';
SumR
esult{
204,1,
m}='
Textur
e_CL12
';
%Sum
Result
{205,1
,m}='Uneve
nTone
';
%Sum
Result
{206,1
,m}='Macro
Textur
e';
%Sum
Result
{207,1
,m}='Micro
Textur
e';
end
138
ifi
scell(
X)==0;
Le
ngthX=1;
else
Le
ngthX=leng
th(X);
end
forin
d=1:
Length
X;
%
ind:現在処理しているファイルの番号
disp
(['run
ning..
...'num2st
r(ind)'of'num
2str(Len
gthX)'
files'])
ifisc
ell(X)==0; %画像を１枚選択
Img=imre
ad(X);
%画像ファイル名の中から拡張子を取り除く
dotw
here=find(
X=='.'
);
Xnam
e=X(
1:dotw
here-1
);
else
%画像を複数選択
Img=imre
ad(X{i
nd});
%画像ファイル名の中から拡張子を取り除く
dotw
here=find(
X{ind}
=='.')
;
Xnam
e=X{
ind}(1
:dotwh
ere-1)
;
end
%画像の種類判定
[M,N,d]=size(I
mg);
cd(Mot
herDir
);
ifd==3%カラー画像
%
画像のグレースケール変換
Im
gGray=
rgb2gr
ay(Img
);
FF
TimgRG
B=ff
t2(dou
ble(Im
g));
FF
TimgGr
ay=f
ft2(do
uble(I
mgGray
));
Sh
iftFFT
imgRGB=fft
shift(
FFTimg
RGB);
Sh
iftFFT
imgR=Shift
FFTimg
RGB(:,:,1)
;
Sh
iftFFT
imgG=Shift
FFTimg
RGB(:,:,2)
;
Sh
iftFFT
imgB=Shift
FFTimg
RGB(:,:,3)
;
Sh
iftFFT
imgGra
y=ff
tshift
(FFTim
gGray)
;
139
FF
TRpowe
r=ab
s(Shif
tFFTim
gR);
FF
TGpowe
r=ab
s(Shif
tFFTim
gG);
FF
TBpowe
r=ab
s(Shif
tFFTim
gB);
FF
TGrayp
ower=abs(S
hiftFF
TimgGr
ay);
%
周波数空間計算
[f
1,f2]
=freqs
pace([
MN],'meshg
rid');%周波数空間を作成
Fr
equenc
ySpace=sqr
t(f1.^
2+f2
.^2); %周波数空間上で原点
からの距離を計算
%
PowerS
pectrumを格納するベクトルを用意、b
=0に対応する値
[i
ndx]=find(
Freque
ncySpa
ce==0);
%
周波数空間で、原
点からの距離が[0]に等しい座標(y,x)を検出
Po
werR=log10
(sum(F
FTRpow
er(ind
x)));
Po
werG=log10
(sum(F
FTGpow
er(ind
x)));
Po
werB=log10
(sum(F
FTBpow
er(ind
x)));
Po
werGra
y=lo
g10(su
m(FFTG
raypow
er(ind
x)));
Po
werSpe
ctrum=zero
s(206,5);
Po
werSpe
ctrum(
1,1)=0;
Po
werSpe
ctrum(
1,2)=Power
R;
Po
werSpe
ctrum(
1,3)=Power
G;
Po
werSpe
ctrum(
1,4)=Power
B;
Po
werSpe
ctrum(
1,5)=Power
Gray;
forb=
0:0.00
5:0.99
5
f=ro
und(b.
*200+2
);
[ind
x]=find
(Frequ
encySp
ace>b&Fr
eq
uencySpa
ce<=(b
+0.005)
);
%周波数空間で、原点からの距離が[b,b+0.005
]に等し
い座標(y
,x)を検出
Po
werR=log10
(sum(F
FTRpow
er(ind
x)));
Po
werG=log10
(sum(F
FTGpow
er(ind
x)));
Po
werB=log10
(sum(F
FTBpow
er(ind
x)));
Po
werGra
y=lo
g10(su
m(FFTG
raypow
er(in
dx)));
Po
werSpe
ctrum(
f,1)=2/((b
+0.005
)*ppm
);
Po
werSpe
ctrum(
f,2)=Power
R;
Po
werSpe
ctrum(
f,3)=Power
G;
Po
werSpe
ctrum(
f,4)=Power
B;
Po
werSpe
ctrum(
f,5)=Power
Gray;
%end
end
fork
=2:5 %Ton
e,Sh
ine,Uneven
Tone,Macr
o-Tex,M
icroTex
140
の計算
Po
werSpe
ctrum(
202,k
)=me
an(Pow
erSpe
ctrum(8:
20,k))
;
Po
werSpe
ctrum(
203,k)=me
an(Pow
erSpe
ctrum(23
:26,k)
);
%P
owerSp
ectrum
(204,k)=m
ean(Po
werSp
ectrum(5
:21,k)
);
%P
owerSp
ectrum
(205,k)=me
an(Pow
erSpe
ctrum(22
:61,k)
);
%P
owerSp
ectrum
(206,k)=
mean(Pow
erSpec
trum(6
2:101,
k));
end
else
ifd=
=1%
グレー画像
Im
gGray=Img;
FF
TimgGr
ay=f
ft2(do
uble(I
mgGray
));
Sh
iftFFT
imgGra
y=ff
tshift
(FFTim
gGray)
;
FF
TPower=abs
(Shift
FFTimg
Gray);
%
周波数空間計算
[f
1,f2]
=freqs
pace([
MN],'meshg
rid');%周波数空間を作成
Fr
equenc
ySpace=sqr
t(f1.^
2+f2
.^2); %周波数空間上で原点
からの距離を計算
%P
owerSp
ectrumを格納する空ベクトルを用意、b=0のデータを入
力
[i
ndx]=find(
Freque
ncySpa
ce==0);
%周波数空間で、原点
からの距離が[0
,0.00
5]に等しい座標(y,x)を検出
Po
werGra
y=lo
g10(su
m(FFTP
ower(i
ndx)))
;
Po
werSpe
ctrum=zero
s(206,2);
Po
werSpe
ctrum(
1,1)=0;
Po
werSpe
ctrum(
1,2)=Power
Gray;
forb=
0:0.00
5:0.99
5
f=round(
b.*200+2);
[ind
x]=find
(Frequ
encySp
ace>b&Fr
eq
uencySpa
ce<=(b
+0.005)
);
%周波数空間で、原点からの距離が[b,b+0.005
]に等し
い座標(y
,x)を検出
PowerG
ray=log10(
sum(FF
TPower
(indx
)));
PowerS
pectru
m(f,1)=2/(
(b+0.0
05)*p
pm);
PowerS
pectru
m(f,2)=Pow
erGray
;
end
141
%To
ne,Sh
ine,U
nevenT
one,M
acro-T
ex,M
icro-Texの計算
Po
werSpe
ctrum(
202,2
)=Po
werSpe
ctrum
(8:20,2)
;
Po
werSpe
ctrum(
203,2)=me
an(Pow
erSpe
ctrum(23
:26,2)
);
%P
owerSp
ectrum
(204,2)=m
ean(Po
werSp
ectrum(5
:21,2)
);
%P
owerSp
ectrum
(205,2)=me
an(Pow
erSpe
ctrum(22
:61,2)
);
%P
owerSp
ectrum
(206,2)=
mean(Pow
erSpec
trum(6
2:101,
2));
end %idd==3/d==1の end
Ba
ndSpli
t(Moth
erDir,Xnam
e,Im
gGray,Shif
tFF
TimgGray
,M,N);
%FFTPowerSpectr
umを csvファイルに保存
%csvw
rite([
'FFT_'Xname'.csv
'],Po
werSpe
ctrum
);
ifd==3
SumRes
ult{1,ind+1
,1}=Xname
;
SumRes
ult{1,ind+1
,2}=Xname
;
SumRes
ult{1,ind+1
,3}=Xname
;
SumRes
ult{1,ind+1
,4}=Xname
;
ifind=
=1
fork=1:20
6
SumR
esult{
k+1,i
nd+1,1}=P
owerSp
ectru
m(k,2);
SumR
esult{
k+1,i
nd+1,2}=P
owerSp
ectru
m(k,3);
SumR
esult{
k+1,i
nd+1,3}=P
owerSp
ectru
m(k,4);
SumR
esult{
k+1,i
nd+1,4}=P
owerSp
ectru
m(k,5);
end
el
se
fork=1:20
6
SumR
esult{
k+1,i
nd+1,1}=P
owerSp
ectru
m(k,2);
SumR
esult{
k+1,i
nd+1,2}=P
owerSp
ectru
m(k,3);
SumR
esult{
k+1,i
nd+1,3}=P
owerSp
ectru
m(k,4);
SumR
esult{
k+1,i
nd+1,4}=P
owerSp
ectru
m(k,5);
end
en
d
else
ifd==
1
SumRes
ult{1,ind+1
,1}=Xname
;
ifind=
=1
142
fork=1:20
6
SumR
esult{
k+1,i
nd+1,1}=P
owerSp
ectru
m(k,2);
end
el
se
fork=1:20
6
SumR
esult{
k+1,i
nd+1,1}=P
owerSp
ectru
m(k,2);
end
en
d
end
end %fori
nd=1
:lengt
h(X)の end
fork=
1:201
SumRes
ult{k+
1,1,1
}=Po
werSpe
ctrum(
k,1);
SumRes
ult{k+
1,1,2
}=Po
werSpe
ctrum(
k,1);
SumRes
ult{k+
1,1,3
}=Po
werSpe
ctrum(
k,1);
SumRes
ult{k+
1,1,4
}=Po
werSpe
ctrum(
k,1);
end
cd
(Mothe
rDir);
ifd==3
SumRes
ult_Re
d=Su
mResul
t(:,:,
1)';%
SumRe
sultの結果を行、
列転置して保存
SumRes
ult_Gr
een=SumRes
ult(:,
:,2)';
SumRes
ult_Bl
ue=S
umResu
lt(:,:
,3)';
SumRes
ult_Gr
ay=S
umResu
lt(:,:
,4)';
disp('
export
ingda
taint
oExce
l.....
.')
xlswri
te('FF
T(4).x
ls',S
umResu
lt_Red
,'RE
D');
xlswri
te('FF
T(4).x
ls',S
umResu
lt_Gre
en,'
GREEN');
xlswri
te('FF
T(4).x
ls',S
umResu
lt_Blu
e,'B
LUE');
xlswri
te('FF
T(4).x
ls',S
umResu
lt_Gra
y,'G
RAY');
el
seifd
==1
SumRes
ult_t=SumR
esult(
:,:,1)
';
disp('
export
ingda
taint
oExce
l.....
.')
xlswri
te('FF
T(4).x
ls',S
umResu
lt_t(:
,:),'GRAY');
en
d
143
disp('
comple
ted')
end
144
APPENDIX C
List of Published Paper, Achievements
Original Paper:
[1] Akira Matsubara, “Differences in the surface and subsurface reflection characteristics of
facial skin by age group”, Skin Research and Technology, vol. 18, pp. 29-32, 2012.
[2] Akira Matsubara, Zhiwu Liang, Yuji Sato and Keiji Uchikawa, “Analysis of human
perception of facial skin radiance by means of image histogram parameters of surface and
subsurface reflections from the skin”, Skin Research and Technology, Article first
published online : 25 AUG 2011
Meetings (Oral Presentation)
“Skin Translucency; what is it and how is it measured?”, International Federation of Society
of Cosmetic Chemists, Oct 2006
“OBJECTIVE MEASUREMENT OF PERCEPTION OF INTANGIBLE SKIN BEAUTY”,
Japan Society of Kansei Engineering, Sept 2011
Reviews
松原晃, “透明感肌の評価法”, フレグランスジャーナル, 1 月号, pp. 61-63, 2007 (Akira
Matsubara, “Skin Translucency: What is it and how is it measured?”, Fragrance Journal,
Issue of January, pp. 61-63, 2007: In Japanese)
145