Article Ouarda
Article Ouarda
Distance Method
Abstract—Human skin detection is an essential phase in face detection. The existing algorithms can be categorized into four
detection and face recognition when using color images. Skin classes: explicit skin classifiers, parametric classifiers, and non-
detection is very challenging because of the differences in parametric classifiers [7].
illumination, differences in photos taken using an assortment of
cameras with their own characteristics, range of skin colors due
II. SKIN COLOR CLASSIFIERS
to different ethnicities, and other variations. Numerous methods
have been used for human skin color detection, including the For the segmentation of skin color pixels, various
Gaussian model, rule-based methods, and artificial neural techniques have been proposed. Some researchers analyzed the
networks. skin color information for skin color detection by converting
In this article, we introduce a technique of using Euclidean into different color space such as RGB, HSV and YCbCr …
distance and thresholding based color image segmentation etc.
algorithm for different skin colors withe the database SFA to
enhance the capabilities of skin detection.
A. Explicit methods
Keywords—Face detection; Skin segmentation; Skin color Different individuals belonging to different ethnic groups,
detection; Euclidean distance. race and region noticeably have different skin color tone. But
under the influence of same lightning conditions, skin color of
I. INTRODUCTION different individuals falls in small area of color space [8, 9].
The explicit classifiers which are the easiest to use and are
Image processing now is one of the most scientific courses, frequently employed, use a threshold strategy to distinguish
which can use in very applicable and industrial projects. Some between skin and non-skin pixels [9]. Basically, they
of these popular projects are face detection [1], human characterize the limits of the skin region by utilizing a set of
tracking, Human identification, visual tracking for surveillance, fixed thresholds. Although such classifiers are direct pixels [2].
hand gesture recognition, searching, image retrieval and So one of the straight forward approach is to explicitly define
filtering image contents on the web and many others. One the the decision boundary of skin color cluster for different color
most stages which should be used in these cases, is skin spaces. For example, the skin is classified into the RGB color
detection. space by using the following rules:
Skin detection is a good example of image segmentation, R > 95 and G > 40 and B > 20 and
which can be accomplished by classifying image pixels as skin Max{R, G, B} – min {R, G, B} > 15 and (1)
and non-skin pixels [2].The importance of skin color detection abs(R-G) > 15 et R > G et R > B
comes from its use as a primary operation in many
applications, such as face detection [3], surveillance systems
[4], hand gesture [5], …etc. For example, face detection is B. Non-Parametric Methods
accomplished by taking out the joint facial characteristics and Non-parametric approaches do not depend on the shape of
by employing skin color detection as a primary step to specify the distribution function of hue. They use color 2D and 3D
the face area. As a result, accurate and fast face detection can histograms representing the distribution (or probability density)
be accomplished. of the skin tone in a color space .The advantage of using color
Many researchers have investigated skin color features. histograms is that they are robust (invariants) to occlusions and
Their results have shown that skin color has a limited range of changes in perspective. They can differentiate a large number
hues and is not deeply saturated [6]. Thus, human skin color is of objects in a recognition task [9, 10].
clustered within a small area in the color space. In the past few
years, several algorithms have been proposed for skin
C. Parametric Methods
Parametric classifiers can be based on a single Gaussian
model, multiple Gaussian clusters, a mixture of Gaussian III. THRESHOLDING METHOD
(MoG) models, or an elliptic boundary model. Generally, the
characterization speed of these classifiers is slow. In fact, they
The main steps of the skin color detection are summarized in
need to process each pixel individually. Additionally, these
the algorithm of figure 1:
methods have low detection accuracy, as they rely on
approximated parameters rather than authentic appropriate skin
colors [8] [9].
This paper is organized as follows. Section 2 presents the
main concepts and work related to skin detection. Section 3
describes the proposed method for adaptive detection of human
skin.
The skin color is a perfect sign to the presence of human
skin in a specific scene and a helpful indicator to minimize the
search area to include the hand and face only. Skin color
modeling techniques have been widely utilized for this
purpose, the selection of color modeling techniques depends on
the nature of the application with the suitable color model [11,
13].
A representative pixel from the skin region can be chosen
as the average pixel. Red, Green and Blue domain intensities of
a skin pixel are chosen as the average pixel intensities.
Euclidean distance is calculated for every image pixel from the
average intensities.
Figure 1. Algorithm skin detection for different color spaces.
In mathematics, the euclidean distance or euclidean metric
is the "ordinary" distance between two points in euclidean SFA human skin image database was created as a tool to
space. With this distance, Euclidean space becomes a metric assist research in computer vision [17]. It is composed of
space. The associated norm is called the Euclidean norm. Older samples of skin, samples of non-skin, the original images and
literature refers to the metric as Pythagorean metric. A the ground truths. The sample images used to build the SFA
generalized term for the euclidean norm is the L2 norm or L2 database were retrieved from two classic and widely used in
distance. the literature face image databases, AR [18] and FERET [19].
These two databases consist of facial color images. Every
The utility of the metrics and distance is to compare the image of SFA was saved in JPEG format, with 100% of quality
similarities and differences between two vectors. This is level.
important, for example, in the field of classification. It is more
likely that two such vectors are in the same class as two In this step, we use for example the HSV (Hue saturation
dissimilar vectors. In fact, the use of metrics to measure value) color space model to consider the luminance effects.
distances is usually an essential step to the automatic Here we must convert our RGB image into its corresponding
classification [14]. HSV color space. After that we need to extract each of these
three (H, S and V) components, which will provide necessary
information for skin color. Then we use only two (H & S)
The image euclidean distance (ED) [15] considers the components to be satisfied against the following equation.
spatial relationship between the pixels of different images and Moreover we can say that a pixel whose H and V components
can easily be embedded in existing image recognition satisfied the (2) will be treated as a pixel in the candidate skin
algorithms that are based on (ED). uses the prior knowledge color region [13].
that pixels located near one another have little variance in gray
scale values, and defines a metric matrix according to the Hence after executing this module we will get an image
spatial distance between pixels. containing only the candidate skin color region as shown in
figure 4. All the skin color regions are extracted:
In 2-D, the Euclidean distance between x and y is:
(2) (3)
The objective of this research is to segment the potential
Conversion according to each area shown in Figure 3 left
skin regions in color images donated despite the change in the
race, background, lighting, etc. In this study, we propose an column.
algorithm for skin segmentation using different color spaces The segmentation procedure is done by scanning all the
and the Euclidean distance to improve accuracy rate. pixels of the image in search of those observing the desired
threshold according to use space. The pixel belonging to the
interval thresholding desire takes "1" if it takes "0". This pixel is a triplet of values namely Red, Green and Blue.
process produces a binary image showing the segments having Segmentation provides better results in RGB color model when
flesh color (white) other not having the skin color will take the compared to other color models. Segmentation in color domain
black [11,12] as shown in Figure 3 column center. is based on similarity detection rather than discontinuity based.
Similarity based detection directly groups the similar pixels.
The algorithm involves selecting an estimate of the average
This step is the most important as well as the most effective
color that is to be segmented. Classifying each RGB pixel in
module in the whole face detection process. In biometry, the
the image as having the specified average color or not. I.e.
image is passport size photo containing only the face region of
similar pixels are grouped together.
a single person. However, if there are some other skin color
regions other than face, it will fail and locate all of them as a After creating a database (skin, not skin) now for each
face region but if we do some morphological operation. For mask (1x1 pixels, 3x3 pixels, 5x5 pixels, 19x19 pixels and
example, a face region contains a larger number of pixels than 35x35 pixels) we will calculate the euclidean distance ED of
the hand, foot etc, provided that there is not so much exposure each pixel, and then compares the differences in intensity pixel
of the body part or so much skin color region which are carrying by pixel. that is to say by calculating the Euclidean distance
larger number of pixels than the face region. Here we use between the components of the pixels (color, brightness) pixel
morphological operation which removes all connected components by pixel by comparing the two images to skin 'D1' and non-
(objects) that have fewer than P pixels, producing another binary skin 'D2' to the three planes then we have extracted the
image as shown in Figure 3 right column. minimum for each case then going to do a comparison between
After getting all the connected components we need to plot a the two distances (D1, D2):
boundary around each of this region. If D1 > D2 Class 1
If D1 < D2 Class 2
IV. EUCLIDEAN DISTANCE METHOD (ED)
After the classification, we have acquired an image segment for
Class “0” black and Class “1” white. the results of
The diagrem of the methode skin detection using the classification shown in figure 4.
euclidien distance in figure 2
V. EXPERIMENTAL RESULTS
(4)
In our method thresholding we use just original images and
ground truths data. However, in distance euclidean we use the
samples (skin and non skin) and the results are presented for
the samples of dimension of 1 pixel, 3x3 pixels, 5x5 pixels,
Figure 2. Algorithm skin detection for euclidean distance (ED). 7x7 pixels 19x19 pixels and 35x35 pixels .
The data skin and non-skin samples were randomly
retrieved considering the GT mask to define which is a skin In the method of thresholding for each space (RGB, HSV,
sample and which is a non-skin sample. From each image it YUV and YCBCR) we adopted a threshold drawn from the
was retrieved three samples of skin and five of non-skin. Each scientific literature. After the execution our programs we
sample has a central pixel, from which other sample sizes were achieved a detection rate is acceptable then [11], for example
created. The samples vary in two pixels, from a sample of 1x1 in RGB space his accuracy that is must not pass 78% but in our
pixel to the bigger one, of 35x35 pixels dimension [17]. work the detection rate over 90%. , the same thing in the HSV
space the detection rate is 35% it is lower for our results very
In the RGB color model, each color appears in its primary between 76% and 94% .same thing for YCBCR space.
spectral components of red, Green and blue. Each RGB color
TABLE1. SKIN DETECTION USING THRESHOLDING METHOD IN DEFFRENT
SPACES.
Method Threcholding
Images
The Spaces RGB (%) HSV (%) YCBCR YUV (%)
(%)
Image 1 94.82 94.29 94.52 87.65
Image 2 95.60 84.07 93.99 92.86
Image 3 93.68 76.07 95.79 96.27
Image 4 97.88 91.46 98.03 92.07
Image 5 95.60 88.95 96.60 87.57
Image 6 82.41 90.77 79.66 89.43
Image 7 94.34 92.81 95.67 88.83
Article [11] 78 36 75 -
TABLE 2. SKIN DETECTION USING EUCLIDIEN DISTANCE (ED) IN RGB SPACE.
REFERENCES
[1] R.-L. Hsu, M. Abdel-Mottaleb, and A. K. Jain, “Face Detection In Color
Images ” IEEE Trans. PAMI, vol. 24, no. 5, pp. 696-707, 2002.
[2] P. Kakumanu, S. Makrogiannis and N. Bourbakis, “A Survey Of Skin-
Color Modeling And Detection Methods” Pattern Recognit. Vol.40,
No 3, pp. 1106-1122, March, 2007.
[3] C. Zhipeng, “Face Detection System Based On Skin Color Model” Int.
Conf Netw. Digit. Soc., pp. 664–667 ,May, 2010.
[4] Z. Zhang, H. Gunes, M. Piccardi, “Head Detection For Video
Surveillance Based On Categorical Hair and Skin Colour Models” IEEE
International 478–485. Conference on Image Processing (ICIP), pp.
1137–1140 2009.
[5] R.Z.Khan, N.A.Ibraheem, “Gesture Algorithms Based on Geometric
Features Extraction and Recognitio” International Journal of Advanced
Research in Computer Science and Software Engineering, pp. 284-290.
Vol 3, No 11, November 2013.
[6] H.K. Al-Mohair, J. Mohamad-Saleh and S.A. Suandi, “Human Skin
Color Detection: A Review” on neural network perspective, Int. J.
Innov.Comput. Inf. Control Vol 8, No 12, pp. 8115–8131, 2012.
[7] M. Abdullah-Al-Wadud, M. Shoyaib, O. Chae, “A Skin Detection
Approach Based On Color Distance Map” EURASIP J. Adv. Signal
Process. Vol 2008 No 199, January2008.
[8] P. PEER, J.KOVAC, and F .SOLINA, “Human Skin Colour Clustering
For Face Detection” In submitted to EUROCON 2003 - International
Conference on Computer as a Tool, Vol 2, pp. 144 - 148, Sept 2003.
[9] A. Kumar and SH. Malhotra, “Pixel-Based Skin Color Classifier: A
Review” International Journal of Signal Processing, Image Processing
and Pattern Recognition, Vol. 8, No.7, pp. 283-290, 2015.
[10] V.Vezhnevets, V. Sazonov and A.Andreeva, “A survey On Pixel-Based
Skin Color Detection Techniques”, Proc. Graph., Vol. 15, pp. 85–92,
2003.
[11] Ch.E .Bencheriet, A.E. Boualleg and H. Tebbikh, “Segmentation de la
Couleur de Peau par Seuillage Selon Différents Espaces de Couleur”,
3èmes Journées Internationales sur l’Informatique Graphique, JIG’2007.
[12] Ch.E .Bencheriet, A.E. Boualleg, H. Tebbikh, B. Guerzize and W.
Belguidoum “Détection de Visages par Méthode Hybride Couleur de
Peau et Template Matching”, SETIT , march 2007.
[13] S. Roy and S. K. Bandyopadhyay, “Face Detection Using a Hybrid
Approach That Combines HSV and RGB”, IJCSMC, Vol. 2, No. 3,
pp.127 – 136, March 2013.
[14] R. FABBRI, L. F. COSTA, J. C. TORELLI and O. M. BRUNO, “2D
Euclidean Distance Transform Algorithms:A Comparative Survey”,
ACM, Vol. 40, No. 1, February 2008.
[15] J. Liand Bao-LiangLu, “An adaptive image Euclidean distance”,Pattern
Recognition Vol 42 ,pp. 349 – 357, 2009.
[16] N.Selvarasu, A. Nachiappan and N.M.Nandhitha, “Euclidean Distance
Based Color Image Segmentation of Abnormality Detection from
Pseudo Color Thermographs”, International Journal of Computer Theory
and Engineering, Vol. 2, No. 4, pp. 1793-8201, August, 2010.
[17] CASATI, J. P. B. ; MORAES, D. R. and RODRIGUES, E. L. L. “SFA:
A Human Skin Image Database Based On FERET and AR Facial
Images”. In: IX Workshop de Visão Computacional, 2013, Rio de
Janeiro. Anais do VIII Workshop de Visão Computacional, 2013.
[18] A. Martinez and R. Benavente, “The AR Face Database,” in CVC o
Technical Report No 24, June 1998.
[19] P. Phillips, H. Moon, P. Rauss, and S. Rizvi, “The FERET Evaluation
Methodology for Eace-Eecognition Elgorithms,” in Computer Vision
and Pattern Recognition. Proceedings. IEEE Computer Society
Conference, Vol 22, No 10, pp. 137–143, 1997.