You are on page 1of 11

西南交通大学学报

第 55 卷 第 1 期 Vol. 55 No. 1
JOURNAL OF SOUTHWEST JIAOTONG UNIVERSITY
2020 年 2 月 Feb. 2020

ISSN: 0258-2724 DOI:10.35741/issn.0258-2724.55.1.17

Research article

Computer and Information Science

IMAGE SEGMENTATION FOR SKIN DETECTION

图像分割,用于皮肤检测

Sajaa G. Mohammed a, Abdulrahman H. Majeed a, Ali Aldujaili b, Enas Kh. Hassan c, Safa S. Abdul-Jabbar d
a
Department of Mathematics, Colleges of Sciences, University of Baghdad
Al-Jadriya, Karrada, Baghdad, Iraq, saj85_gh@yahoo.com, ahmajeed6@yahoo.com
b
Dormitory, University of Baghdad
Al-Jadriya, Karrada, Baghdad, Iraq, ali@uobaghdad.edu.iq,
c
Department of Computer Science, Colleges of Sciences, University of Baghdad
Al-Jadriya, Karrada, Baghdad, Iraq, enas.mkhazal@gmail.com
d
Department of Computer Science, Colleges of Sciences for Woman, University of Baghdad
Al-Jadriya, Karrada, Baghdad, Iraq, safa.s@csw.uobaghdad.edu.iq

Abstract
Human skin detection, which usually performed before image processing, is the method of discovering
skin-colored pixels and regions that may be of human faces or limbs in videos or photos. Many computer
vision approaches have been developed for skin detection. A skin detector usually transforms a given
pixel into a suitable color space and then uses a skin classifier to mark the pixel as a skin or a non-skin
pixel. A skin classifier explains the decision boundary of the class of a skin color in the color space based
on skin-colored pixels. The purpose of this research is to build a skin detection system that will
distinguish between skin and non-skin pixels in colored still pictures. This performed by introducing a
metric that measures the distances of pixel colors to skin tones. Results showed that the YCbCr color
space performed better skin pixel detection than regular Red Green Blue images due to its isolation of the
overall energy of an image in the luminance band. The RGB color space poorly classified images with
wooden backgrounds or objects. Then, a histogram-based image segmentation scheme utilized to
distinguish between the skin and non-skin pixels. The need for a compact skin model representation
should stimulate the development of parametric models of skin detection, which is a future research
direction.

Keywords: Skin Detection, Image Segmentation, Skin-Colored Pixel, Color Space

摘要 通常在图像处理之前执行的人类皮肤检测是在视频或照片中发现可能是人脸或四肢皮肤颜色
的像素和区域的方法。已经开发出许多用于皮肤检测的计算机视觉方法。皮肤检测器通常将给定
像素转换为合适的颜色空间,然后使用皮肤分类器将像素标记为皮肤或非皮肤像素。肤色分类器
根据肤色像素解释颜色空间中肤色类别的决策边界。这项研究的目的是建立一种皮肤检测系统,
该系统可以区分彩色静态图片中的皮肤像素和非皮肤像素。通过引入一种度量标准来执行此操作
2 Mohammed et al. / Journal of Southwest Jiaotong University / Vol. 55 No. 1 Feb. 2020

,该度量标准可测量像素颜色与肤色的距离。结果表明,由于六价铬色彩空间隔离了亮度带中图
像的整体能量,因此它比常规的红绿蓝图像执行了更好的皮肤像素检测。RGB 颜色空间对具有木
制背景或对象的图像分类不佳。然后,基于直方图的图像分割方案可用来区分皮肤像素和非皮肤
像素。对紧凑型皮肤模型表示的需求将刺激皮肤检测参数模型的发展,这是未来的研究方向。

关键词: 皮肤检测,图像分割,肤色像素,色彩空间

I. INTRODUCTION [5]. Most published research on skin discovery


In the past decade, a large number of skin reports a real discovery rate of about 95% and
discovery applications have been developed to false discoveries ranging between 15% and 30%,
improve the human–computer interaction with irrespective of how the color space is selected
new, nontraditional interfaces. Different and categorized [6]. The discovery of skin
techniques have been developed using various indicates the presence of a human skin in a digital
features that represent the human face. In this image by converting the original image to a dual
work, we focus on the color of the skin. Color is image in which “1” represents a skin pixel and
the most important feature that we can use to “0” represents a non-skin pixel. Development of
distinguish between the face and background new techniques, such as for face discovery, hand
regions in an image of the head and shoulders of gesture detection, and semantic filtering of web
a human. Other parts of the body than the face content, is an interesting problem and an
can have the same skin tone when observed out important pre-processing step [7].
of their whole context. Nevertheless, the number
of false positives can be reduced by searching II. LITERATURE SURVEY
only image regions that may have face pixels. There has been considerable work on skin
Skin color filtering is a significant procedure for detection. The related studies below are
removing non-face pixels. Skin color filters can important reviews.
swiftly remove non-skin-colored pixels, thereby 1. In 2018, Nguyen-Trang et al.
reducing the search region for the face skin investigated and evaluated a new method for
detection step. In skin detection, skin color detecting skin in colored images. The proposed
information is collected, and colors are method is competitive in terms of detection rate
segmented through a preparation process to and outperformed others in terms of false positive
obtain superior detection performance. After skin rate and accuracy [7].
color detection, the skin color of the detected 2. In 2018, Li et al. proposed a method of
region is searched for a face candidate. mitigating the erlier mentioned deficiencies; that
Morphological operations are used to remove is, the parameters in the transformation matrix
noise from images [1]. The skin plays an from the RGB space to the YCbCr color space
important role in digital image processing, and were changed after the transition matrix was
discovering the color of human skin is important modified. Consequently, the face detection
in many applications. Any image may have many system can detect human faces quickly and
objects this whole research is based on detecting accurately, and the detection rate is improved [8].
the human objects with skin from non human 3. In 2017, Yusuf et al. proposed a study
objects [2]. The first step in face feature that is used in various fields of application in
extraction and recognition is face detection. The daily life. Each color model contains a specific
main purpose of face detection is to determine representation space and components and can
whether a face is present in an image or not; if convert from one color space to another using a
one is present, the location of the face in the standard formula. They concluded that the
image will be returned [3]. Many of the selection of a proper color model for a specific
techniques for this application assume that the application depends on the characteristics of the
location of the face is defined and ready for the model and the nature of the application. A
following step. The first problem in face framework for efficient face detection using skin
detection is selecting a proper color model for color segmentation was suggested, and a
skin color segmentation [4]. Standard skin color discovery rate of up to 97.22% was acquired by
detection techniques are negatively affected by using a standard database [3].
changing lighting conditions, complex
backgrounds, and surfaces with skin-like colors
3

III. PROBLEM STATEMENT AND boundaries between different objects or areas.


Segmentation based on similarity, on the other
MOTIVATIONS hand, relies on identifying regions in the input
The main problems could be summarized in
image that are similar according to some
the following points.
predetermined standards.
1. To what extent can skin color be
Although many image segmentation
modeled with a physics-based approach for
approaches have been proposed, the most popular
several types of skin when viewed by a camera
ones rely on neural networks, while one-
under a range of single and mixed light sources?
dimensional signal segmentation, multi-scale
How accurate is the required light source model?
segmentation, model-based segmentation,
2. According to the dependence of skins
graphic partitioning, region growing and K-mean
appearances one they lightening' color, how' goods
clustering segmentation methods also yield very
and' under' what' conditions' may' they lightening'
good results [11].
colors bee estimated' from' skins given' ac camera'
inputs image.
3. How good a region of skin might bee V. PROPOSED IMAGE IN-PAINTING
distinguished from other skin like objects under'" SYSTEM (IIS)
varying lighting cases using a physics-based' The segmentation scheme adopted in this
approach. work consists of several skin color models, the
4. The reflection properties of the near first of which generates all possible images. The
infrared spectrum can help distinguishes between' second Choosing suitable color space method
skins and other skin like objects. step is the application of the images (checker),
this system performance have been tested. The
IV. IMAGE SEGMENTATIONS RGB (Red Green Blue) is Color space (RGB) for
The process by which an image is partitioned detecting skin color pixels. There'd Convert the
into regions is called segmentation. Several color image from RGB to YCbCr. Fourth Skin
segmentation methods have been proposed, such color detection and the output better skin
as color models and above-mentioned RGB and detection.
YCbCr. Image segmentation based on histogram The full skin color system has been designed
is based on color and is considered suitable for and implemented and its results have been
processing the input image in order to identify analyzed. Standard color image data sets were
and localize multiple faces in skin tone-color used as test materials to investigate the
images. Segmentation method proposed by [9] is performance of the suggested skin color system;
based on human skin chromaticity. The aim is to the results indicate that the efficiency of the
partition an image based on immediate changes proposed scheme is encouraging compared to
in intensity, which tend to occur on the other state-of-the-art skin detection schemes.

Figure 1. The proposed system skin color

The layout of proposed skin color system


consists of two modules. Each module has its
own method. Figure 2 shows the system layout.
4 Mohammed et al. / Journal of Southwest Jiaotong University / Vol. 55 No. 1 Feb. 2020

Figure 2. The block diagram of proposed system layout

VI. SKIN COLOR MODELING AND Skin color is useful for the detecting a skin of
any color, recognition, and tracking. It has been
DETECTION proven that using skin color information before
Many skin color models, discoveries, and
tested photograph images processing can reduce
methods of segmentation have been proposed in
difficulties and problems. But because of the
the literatures– more than 120 are cited in this
effects of race, environmental lighting, and
section. There' are' many possible strategies for'
background interference, it is difficult to
categorizing these methods, statistical-based to
determine human skin color accurately [14].
physics-based to the used color space to how they
are tested and rated. Given a skin color cue, i.e.,
A. Color Spaces, Skin Modeling, and
using skin as a feature, we can divide the process
Detection
up into three problems:
Skin color is an important feature of human
 choosing a color space; faces and is a useful property in tracking a face,
 modeling the skin color division; and as color processes much faster than any other
 making the actual division, e.g., pixel- feature for facial identification purposes.
based or taking into account the locational Although a human observer can easily identify
relationship. skin color, the task is more difficult for a
Although the third problem is important for machine. In computer vision, it is important to
skin color cues, this sections – due to the overall choose the right color space, as this can refine the
emphasis off this dissertation – focuses on the distribution of skin color in an image. Much of
first two problems, which are prerequisites for the research in skin color and facial localization
the third. A new pixel-based skin color and detection is based on RGB, YCbCr, and hue
identification technique by Vezhnevets, Sazonov, intensity saturation (HIS) color spaces [1], [18].
and' Andreeva defined the following categories In this project, a pixel-based skin discovery
for' modeling skin color divisions: The skin strategy was adopted. This process classifies each
region is clearly defined according to three individual pixel as skin or non-skin, according to
models: nonparametric skin distribution, color value and independent of' its neighbors. The
borderline skin distribution, and dynamic skin core of this strategy is the selection of color space
distribution [12]. Human skin color has a and skin tone modeling. In this dissertation, we
restricted range of hues and is not deeply select YCbCr to represent each pixel because the
saturated because the superficial color of skins is color transformation from RGB to YCbCr is
formed by a combination of blood (red) and simple. More importantly, YCbCr can separate
melanin (brown or yellow). Therefore, the color luminance and chrominance components
of human skin doesn’t fall arbitrarily in a given explicitly. The effect of color transformation on
color space but rather clusters in a small area skin detection performance must be measured
within that space. But this tendency does not with two color space transformations: RGB and
apply to all color spaces. There are plenty of YCbCr. These color models are used in color
color spaces in image processing and varity of image processing and contain both variant and
these have been used to figure out what is the invariant properties with reference to imaging
most suited [13]. conditions. With the color space transformation,
the objective is to decrease the overlap between
5

skin and non skin pixels thereby maximizing of the respective NTSC and PAL systems. These
classification performance. The equations for RGB (Red-Green-Blue) color spaces separate the
color space conversion don't produce values in luminance and coloring information and are
the rate 0–255 always. In our framework the useful for compression applications, however
values are not adjusted We provide a color specifications are somewhat easy. The
comprehensive evaluation of color based skin YCbCr color space is commonly used in image
discovery in six color spaces (RGB, YCbCr) with processing as the luminance in component Y is
nine skin color modeling approaches [13]. separated from the morphology described by the
Cr and Cb components. There are several
B. RGB Color Space for Skin Detection definitions of this transformation. Luminance Y
RGB model simplifies computer graphic is created as a weighted sum of RGB elements
system designs but it's not suitable for all and Cb and Cr elements acquired by deducting Y
applications. The components of red, green and from blue and red RGB elements respectively
blue color are closely related. RGB color space is [1].
the most used color space for digital images. It 1. Y = 0.299 R + 0.587 G + 0.11 B;
encodes colors as an additional mix of three 2. Cb = B – Y;
primary colors: green (G), blue (B) and reds (R1). 3. Cr = R – Y.
RGB color space is often visualized as a 3D cube Cb and Cr materials are used to distinguish
where R, G and B are the three vertical axes with skin color information. Below the conversion
red, green, and blue at the corners on each axis. formats from RGB to YCbCr are shown:
Black is its original place. And white at the other 1. Y = 0.299 R + 0.587 G + 0.114 B;
end of the cube. The gray scale follows the line 2. Cb = - 0.168 R – 0.332 G + 0.500 B;
from black to white. e.g. in a 24-bit color 3. Cr = 0.500 R – 0.419 G – 0.081 B.
graphics system with 8-bits per color channel, red
is (255, 0, 0) and on the color cube, it is (1, 0, 0) VII. SKIN COLOR SEGMENTATION
[1].
RGB color space is the most used color space
SYSTEM DESCRIPTION
In this project, two color models were used for
for digital image. It encodes colors as an
the segmentation of skin color or discovery of
additional mix of three primary colors: red (R),
skin pixels. These are the RGB and YCbCr color
green (G) and blue (B). One advantage of RGB
models.
space is it's simplicity. However, they are not
uniform, which means that the distances in RGB
A. RGB Skin Detection Implementation
space are not linearly compatible with human
RGB color area is used with a wide range and
perception. Moreover, RGB color space doesn't
the default color space for storing and
separate the lighting and chrome, and the
representing digital images is normal. We can
component R, G and B are closely related. RGB
obtain any color space from non-linear or linear
pixel brightness is a linear combination of the
or reformation of RGB. The RGB color space is
values of R, G, and B. So, changing the skin
used by graphics cards, monitors or LCDs and
correction brightness affects all component of R,
computers [14]. The color space (RGB) for the
G, and B [13].
discovered skin color pixels is determined
whether a specific pixel of the input image
C. YCbCr Color Space for Skin Detection
corresponds to the skin color or not with the help
In the YCbCr color model as well, the
of guidelines. It was noted that the original face
distribution of skin areas is consistently across
discovering algorithm was developed to work
different races in the areas of' Cb and Cr colors,
better under standard daylight illumination (CIE
and the RGB color models issue lighting
illuminant D65). After the skin color
sensitives, thus, when we use differently color
categorization is done for every pixel of the
models in uncontrolled conditions, and we reach
picture, the skin region segmentation takes place.
a result for skin color discovery [4].
Unsuitable regions are then eliminated on the
The YCbCr color area has been defined in
basis of the geometric properties of the face. The
response to the increasing demands of digital
remaining areas represent faces and exemplify
algorithms in handling video information and has
the described process. Skin color, for example,
become a widely used model for digital video
fair skin is defined as the computer screen that
[17]. It belongs to the family of colors TV
looks like a valuable palette. Above the picture
broadcast spaces. The family includes others like
the round opening is used for the digital camera
YUV and YIQ. YCbCr issue an digital color
lens.
scheme, while YIQ and YUV are representative
6 Mohammed et al. / Journal of Southwest Jiaotong University / Vol. 55 No. 1 Feb. 2020

With the following rules which describe the model. For each pixel in samples of skin, the
skin group in the RGB color space [15]: values of Cb and Cr are calculated. After the Cb
The color of the skin at unified daylight Cr skin color model is built, it can be used for the
illumination ⇒ discovery of skin. The first step in skin detection
 R > 95 AND G > 40 AND B > 20 AND is pixel-based skin discovery, where the skin
max {R, G, B} − min{R, G, B} > 15 AND RGB discoverer start to test each pixels of the input
components must not be close together grayness picture and computes its Cb and Cr. If Cb and Cr
eliminate values of the pixels satisfy the inequalities, then
|R – G| > 15 AND R and G components must this pixel is considered skin. The range of values
not be close together, otherwise we area not for skin color were according to experimental
dealing within the fair complexion results [13].
R > G AND R > B (i.e., R component must be 1. 137 < Cr < 177;
the greatest component) 2. 77 < Cb < 127;
OR 3. 190 < Cb + 0.6 Cr < 215.
The skin color under flashlight or (light) The YCbCr space was chosen for following
daylight lateral illumination ⇒ reasons [1]:
 R > 220 AND G > 210 AND B > 170 1. Bitmaps use RGB planes straight to
AND represent color images. But after conducting some
|R – G| ≤ 15 AND (i.e., R and G components medical research, it proved that the human eye has
must be close together) a different sensitivity to color and brightness.
R > B AND G > B (i.e., B component must be Then they suggested converting RGB to YCbCr.
the smallest component). 2. The luminance component (Y) of YCbCr
This is how we get the basic information is independent of the color, so it can be adopted
about the candidates' faces. So describe the skin to solve the illumination variation problem, and it
color and describe the step [15]. is easy to program.
3. The bunch of skin color is more compact
B. YCbCr Skin Detection Implementation in YCbCr than in the space of other colors.
YCbCr is an encoded non-linear RGB signal, 4. YCbCr has the smallest interference
used in European TV studios and also used for between skin and non-skin data under different
the compression of images. YcbCr is a colorful lighting conditions. YCbCr is used in video
space commonly used in digital video. The compression standards (such as MPEG and
representation makes it easy to get rid of JPEG).
redundant color information [14]. Here, to 5. YCbCr1c is' ac family's off colors spaces
segment the skin color, we choose the YCbCr used' in systems of video. YCbCr was defined for
color space, which is similar to human visual standard-definitions TV used ink they ITU-R
perception in principle. These components of Y, BT.601 standards ford used within digitals
Cr and Cb can be acquired by linear components video.
transformation of RGB color space [16]. The 6. YcbCr is one of two main spaces of colors
traditional YCbCr color space is optimized for used to characterize digital constituent video (the
face detection, and the enhanced color space can other is RGB).
reflect the entire face color, so the system can 7. The difference between YcbCr and RGBs
detect the face more precisely [8]. is that YcbCr represents color ask brightness and
In our work, we build a skin tone model based two color differences signals, and RGB represents
on clearly defined skin areas in the Cb Cr color colors as red, green and blue.
space. The Cb Cr color space has been chosen in
our work for several reasons: It does not contain VIII. EXPERIMENTAL RESULTS
information about the luminance that gives a We implemented the considered skin
general pattern of skin tones; it contains only two segmentation algorithm with Visual Basic 6. For'
components which help to speed up the evaluating the performance of the discovery, we
calculations; and the transformations from RGB conducted many tests on different images,
color space into Cb Cr Proceedings of the including 6 pictures under different' illumination
Geometric Modeling and picturing color space is conditions. Image 1 represents the normal
done by using simple and fast transformations. lighting, Image 2 denotes lighting under sidelight,
The major reason for using clearly defined skin and Image 3 comprises images in a gloomy scene,
areas in building a skin detector is its speed in see Figure 3. Despite the selection of the
detecting skin areas. We have collected human categorization method and the color space,
skin samples from several races to build the published research on skin discovery indicates a
7

real discovery of 95% while false detection rates finding biometrics for facial skin. And skin color
range from 15% to 30% [6]. The database is a distinctive feature of human faces. In a
contains information on many eyes with partial controlled environment, skin discovery can be
closure and round face in each group. The sufficient to identify faces in images. Because
experiment is a results of different lights and the color processing is much faster than other facial
test results in the partial rotation of the face; features, it can be used as the primary process for
moreover, some skin images are revealed. other face detection techniques. Skin detection
However, it is not possible to pass the matching has also been used to locate the ends of the body,
of the eye template or the examination of pre- such as hands.
rules. Human face discovery is the first step in

Figure 3. Original tested images

A. Results of Skin Color Based Discovery in results also confirm that the false detection rate is
RGB Color Space very high, thus causing very low accuracy in
The results from this' experiment reveal that detecting skin over all the body. See Figure 4a
RGB color space is not compatible with skin and Table 1.
discovery, based on skin color categorization. Its

Table 1.
Results of RGB elapsed processing time for both slower and faster image programming techniques
Image name Elapsed time faster Elapsed time slow
1/1000 sec. 1/1000 sec.
Child 1 190 2880
Picture 2 150 2140
Hand 1 190 2920
Picture 5 230 3340
Pepol 360 5460
Child 2 400 5600

B. Results of Skin Color-Based Discovery in false discovery rate is high in RGB, thus causing
YCbCr Color Spaces low accuracy in detecting the skin region. The
Other experiments have been performed on quality of input image is not very good, but the
the RGB color space, which' provided faster and skin model has detected all skin regions with
better' than the results from the YCbCr color little deviation with respect to the expected result
space. The following listed results show that the (Figure 4b and Table 2).
8 Mohammed et al. / Journal of Southwest Jiaotong University / Vol. 55 No. 1 Feb. 2020

b
Figure 4. a - Results of RGB skin detection, b - Results of YCbCr skin detection

Table 2.
Results of YCbCr elapsed processing time for both slower and faster image programming techniques
Image name Time elapsed faster Time elapsed slow
1/1000 sec. 1/1000 sec.
Child 1 470 2190
Picture 2 630 2410
Hand 1 730 2970
Picture 5 800 3370
Pepol 1170 5540
Child 2 1190 5590
According to that we assume that the skin
color is determined by its Cr and Cb color IX. CONCLUSIONS
components, so the luminance component Digital detection of skin in videos and
depends on the light conditions. There is also a pictures is a good way to identify skin color
hypothesis that says the skin color of many pixels, which indicate the presence of faces and
people as well as the types of human skin color hands. All beings in the world have color in their
are the same in the components of the dye. The skin that can look like some types of leather,
difference is according to the luminance wood, fur, etc., which can be mistakenly detected
parameter change, and this indicates that there is by the skin detector. The digital identification of
no need to track the function of the luminance skin is very important and useful in finding
component in detecting skin tone in the YCbCr human hands and faces in controlled
color space. The advantage of this method is environments where it is ensured that the
simplicity and low computational costs. In this background does not contain skin color. Because
project, it is difficult to distinguish between the skin detection depends on locating pixels that
background color of the image if it is similar to contain skin tone, its use is limited to color
the skin color (red). images, that is, it's not useful in gray scales,
9

infrared or other types of image methods that [2] VARMA, S.L. and BEHERA, V.
don't have color. There has been extensive (2017) Human Skin Detection Using
research on finding human faces in photos and Histogram Processing and Gaussian Mixture
videos using other characteristics like local facial Model Based on Color Spaces. In:
features or comprehensive face templates. Skin Proceedings of the International Conference
detection can also be used as an effective pre-
on Intelligent Sustainable Systems,
treatment filter to find potential skin areas in
color images before applying mathematically Palladam, December 2017. Piscataway, New
more expensive skin, hand, or face detectors. Jersey: Institute of Electrical and Electronics
Color is a useful piece of information to discover Engineers, pp. 116-120.
skin. Skin detection is the most common and first [3] YUSUF, A.A., MOHAMAD, F.S.,
applied method for discovering skin color and SUFYANU, Z. (2017) Human Face
meaningfully. The detection of skin color may Detection Using Skin Color Segmentation
avoid a comprehensive search for faces in a and Watershed Algorithm. American Journal
complete picture. Here, we explain how non skin of Artificial Intelligence, 1 (1), pp. 29-35.
color is rejected from an image so that the image [4] RAGHUVANSHI, D.S. and
may only contain skin-like areas, which result in AGRAWAL, D. (2012) Human Face
segmented skin color images for further
Detection by Using Skin Color
processing. For different types of color models,
in the YCbCr color model, the distribution of the Segmentation, Face Features and Regions
areas of skin is consistent across different Properties. International Journal of
generations in the color areas of Cb and Cr; the Computer Applications, 38 (9), pp. 14-17.
RGB color model is sensitive to lighting, thus, [5] KHAN, R., HANBURY, A.,
when we use different models of color in STÖTTINGER, J., and BAIS, A. (2012)
undisciplined conditions, we get a different result. Color based skin classification. Pattern
The accuracy of skin detection depends on both Recognition Letters, 33, pp. 157-163.
the color model and the method of classification [6] ELGAMMAL, A., MUANG, C., and
or detection of skin pixels. Thus, the challenge is HU, D. (2009) Skin Detection - A Short
how to choose appropriate color models for skin Tutorial. Berlin, Heidelberg: Springer.
pixel ratings under different conditions.
[7] NGUYEN-TRANG, T. (2018) A
New Efficient Approach to Detect Skin in
X. SUGGESTIONS FOR FUTURE Color Image Using Bayesian Classifier and
WORKS Connected Component Algorithm.
Some notes about future work are summarized Mathematical Problems in Engineering,
as follows: 2018, 5754604.
1. A combination of color models' overcom
[8] LI, E. and XU, Y. (2018) Face
es all changes
in lighting conditions, and changes in lighting giv Detection Based on Improved Color Space of
es better results than the individual color model. YCbCr. IOP Conference Series: Materials
2. Improvements in skin detection can be Science and Engineering, 439, 032075.
made by scanning skin candidates using effective [9] CHANDRAPPA, D.N.,
classification methods. RAVISHANKAR, M., and RAMESHBABU,
3. After detecting human faces in images, D.R. (2011) Face Detection in Color Images
there should be some additional features that are Using Skin Color Model Algorithm Based on
studied in future work, such as eyes and nose Skin Color Information. In: Proceedings of
detection in faces. A skin region can, however, be the 3rd International Conference on
extracted from a noisy image by interpolating Electronics Computer Technology,
color information. We can also identify skin by
Kanyakumari, April 2011. Piscataway, New
using template matching.
Jersey: Institute of Electrical and Electronics
Engineers, pp. 254-258.
REFERENCES
[10] GONZALEZ, R.C. (2002) Digital
Image Processing. Upper Saddle River, New
[1] POWAR, V., JAHAGIRDAR, A., Jersey: Prentice-Hall.
and SIRSIKAR, S. (2012) Skin Detection in [11] HO, P.-G.P. (2011) Image
YCbCr Color Space. International Journal of Segmentation. Rijeka: InTech.
Computer Applications, 5, pp. 1-4.
10 Mohammed et al. / Journal of Southwest Jiaotong University / Vol. 55 No. 1 Feb. 2020

[12] STORRING, M. (2004) Computer


Vision and Human Skin Color. PhD. thesis, 参考文:
Aalborg University.
[13] YI, Y., QU, D., and XU, F. (2009) [1] POWAR,V.,JAHAGIRDAR,A. 和
Face detection method based on skin color SIRSIKAR,S.(2012)六价铬颜色空间
segmentation and eyes verification. In: 中的皮肤检测。国际计算机应用杂志,5,
Proceedings of the International Conference 第 1-4 页。
on Artificial Intelligence and Computational [2] VARMA , S.L. 和 BEHERA , V.
Intelligence, Shanghai, November 2009.
(2017)使用直方图处理和基于色彩空间
Piscataway, New Jersey: Institute of
Electrical and Electronics Engineers, pp. 的高斯混合模型进行人体皮肤检测。于:
495-501. 2017 年 12 月在帕拉丹举行的国际智能可
[14] KOLKUR, S., KALBANDE, D., 持续系统会议论文集。新泽西州皮斯卡特
SHIMPI, P., BAPAT, C., and JATAKIA, J. 维:电气与电子工程师协会,第 116-120
(2016) Human Skin Detection Using RGB,
页。
HSV and YCbCr Color Models. In: IYER,
B., NALBALWAR, S., and PAWADE, R. [3] A.A. YUSUF , F.S 。 MOHAMAD 和
(eds.) ICCASP/ICMMD-2016. Advances in SUFYANU,Z。(2017)使用肤色分割
Intelligent Systems Research, Vol. 137. 和分水岭算法进行人脸检测。美国人工智
Atlantis Press, pp. 324-332. 能杂志,1(1),第 29-35 页。
[15] KOVAC, J., PEER, P., and SOLINA, [4] RAGHUVANSHI,D.S. 和 AGRAWAL,
F. (2003) Human Skin Color Clustering for D.(2012)通过使用肤色分割,面部特征
Face Detection. In: Proceedings of the IEEE
和区域特性进行人脸检测。国际计算机应
Region 8 EUROCON 2003. Computer as a
Tool, Ljubljana, September 2003. 用杂志,38(9),第 14-17 页。
Piscataway, New Jersey: Institute of [5] KHAN , R. , HANBURY , A. ,
Electrical and Electronics Engineers, pp. STÖTTINGER,J. 和 BAIS,A.(2012)
144-148. 基于颜色的皮肤分类。模式识别字母,33,
[16] CHANG, L. (2017) The Research of 第 157-163 页。
Face Detection Method Based on Adaboost
[6] ELGAMMAL , A. , MUANG , C. 和
Algorithm and Skin Color Segmentation. In:
Proceedings of the IEEE International HU,D.(2009)皮肤检测-简短教程。柏
Conference on Imaging Systems and 林,海德堡:施普林格。
Techniques, Beijing, October 2017. [7] NGUYEN-TRANG,T.(2018)一种使
Piscataway, New Jersey: Institute of 用贝叶斯分类器和连通分量算法在彩色图
Electrical and Electronics Engineers. 像中检测皮肤的新有效方法。工程数学问
[17] BODRYSHEV, V.V., ABASHEV,
题,2018,5754604。
V.M., KORZHOV, N.P., and TARASENKO,
O.S. (2018) Study of Interaction of a Jet with [8] LI,E. 和 XU,Y.(2018)基于改进的
Supersonic Flow on the Parameter of the 六价铬色彩空间的人脸检测。眼压会议系
Intensity of the Video Frame Image. 列:材料科学与工程,439,032075。
Periódico Tchê Química, Special Issue 1, pp. [9] D.N. CHANDRAPPA , M 。
348-358. RAVISHANKAR 和 D.R. RAMESHBABU
[18] ABDULLATIF, A.A., (2011)使用基于肤色信息的肤色模型算
ABDULLATIF, F.A., and AL SAFAR, A.
法在彩色图像中进行人脸检测。于:2011
(2019) Multi-Focus Image Fusion Based on
Stationary Wavelet Transform and PCA on 年 4 月在宽 aku 举行的第三届电子计算机
YCBCR Color Space. Journal of Southwest 技术国际会议论文集。新泽西州皮斯卡塔
Jiaotong University, 54 (5). Available from 维:电气与电子工程师协会,第 254-258
http://jsju.org/index.php/journal/article/view/ 页。
388.
11

[10] 冈萨雷斯,加拿大(2002)数字图像 西 南 交 通 大 学 学 报 , 54 ( 5 ) 。 可 从
处理。新泽西上萨德尔河:普伦蒂斯·霍 http://jsju.org/index.php/journal/article/view/
尔。 388 获得。
[11] HO,P.-G.P。(2011)图像分割。力
耶卡:科技。
[12] STORRING,M.(2004)计算机视觉
与人类肤色。博士论文,奥尔堡大学。
[13] YI,Y.,QU,D. 和 XU,F.(2009)
基于肤色分割和眼睛验证的面部检测方法。
于:2009 年 11 月在上海召开的国际人工
智能和计算智能国际会议论文集。新泽西
州皮斯卡塔维:电气与电子工程师学会,
第 495-501 页。
[14] KOLKUR , S., KALBANDE ,D. ,
SHIMPI,P.,BAPAT,C.和 JATAKIA,J.
(2016)使用 RGB,单纯疱疹病毒和六价
铬颜色模型进行人体皮肤检测。于:
IYER , B. , NALBALWAR , S. 和
PAWADE,R.(编。)联会/国际医疗器
械展览会-2016。智能系统研究进展,第 1
卷。137.亚特兰蒂斯出版社,第 324-332
页。
[15] KOVAC,J.,PEER,P. 和 SOLINA,
F.(2003)用于面部检测的人类皮肤颜色
聚类。在:电气工程师学会地区 8 欧康
2003 的会议记录中。计算机作为工具,卢
布尔雅那,2003 年 9 月。新泽西州皮斯卡
塔维:电气与电子工程师协会,第 144-
148 页。
[16] CHANG,L.(2017)基于 Adaboost
算法和肤色分割的人脸检测方法研究。
于:2017 年 10 月,北京,电气工程师学
会国际成像系统和技术会议论文集。新泽
西州,皮斯卡塔维:电气与电子工程师学
会。
[17] V.V. BODRYSHEV , V.M 。
ABASHEV , N.P 。 KORZHOV 和 O.S.
TARASENKO。(2018)研究喷射流与超
声流在视频帧图像强度参数上的相互作用。
佩里科迪科奇奎米卡,特刊 1,第 348-
358 页。
[18] ABDULLATIF , A.A. ,
ABDULLATIF,F.A,和 AL SAFAR,A。
(2019),基于 YCBCR 色彩空间上基于
平稳小波变换和 PCA 的多焦点图像融合。

You might also like