Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications

Volume: 03 Issue: 02 December 2014, Page No. 46-48

ISSN: 2278-2419

Human Skin Detection from Image using Gaussian

Algorithm
A.Thomas1, M.S.Josephine2, V.Jeyabalaraja3
1
Research Scholar, Department of Computer Applications, Dr. M.G.R. Educational and Research Institute University, Chennai
2
Professor, Department of Computer Applications, Dr. M.G.R. Educational and Research Institute University, Chennai
3
Professor, Department of computer Science Engineering, Velammal Engineering College, Chennai
E-mail: [email protected], [email protected], [email protected]

Abstract-This paper presents the procedures of Human skin typically used as a preprocessing step to find the regions that
detection. For this we propose preprocessing (RGB mean potentially have human images. By detecting skin-colored
value) that combines a smoothed 2-D histogram [4] and regions was used to identify nude pictures on the internet for
GAUSSIAN approach, for automatic human skin detection in the sake of content filtering [2]. Several computer vision
color images. The main objective of this paper is to provide the approaches have been developed for skin detection. A skin
maximum prefect area of human skin to be detected. The detector typically transforms a given pixel into an appropriate
experimental result has show that the performance of the color space and then uses a skin categorizer to identify the
Gaussian approach works well over the Human skin detection, pixel whether it is a skin or a non-skin pixel.
eye detector.
Keywords: Skin detection; 2 d histogram; Gaussian model; II. REVIEW OF LITERATURE
preprocessing RGB mean value; adaptive threshold segmenta-
The basic concept of the image processing and the skin
tion
detection are covered in the review of literature. Also in the
I. INTRODUCTION previous research they already made the skin detection using
MATLAB [1], fusion approach etc. But in this paper, we are
Skin detection is the process of Finding skin-colored pixels and
implemented the Gaussian Algorithm.
regions in an image [1]. This process is
Table: 1-Here is a table to describe the Literature survey for this skin detection.
S.NO TITLE AUTHOR WORKING
Algorithm Fusion for Face It presents the fusion task application using
M. MILGRAM
1 Localization detectors like skin color, auto associative multi
layer perception and ellipse Hough transform.

Fan Hai Xiang, Shahrel

Fusion of multicolor space for Azmin Suandi Identifies the skin color segmentation based on
2
human skin region segmentation color space models.

Recognize human face by low level fusion namely

M. K. Bhowmika, G.
data fusion and high level fusion namely decision
Next level of data fusion for Majumdera, D.
fusion.
3 human face recognition Bhattacharjeeb, D. K. Basub,
M. Nasipurib 1.Data fusion using wavelet decomposition and re-
construction techniques.
2.Decision fusion based on Bayesian formulation
Baozhu Wang1∗, Xiuying Identifies face segmentation in image by skin
Skin detection and segmentation Chang2∗, Cuixiang Liu1
4 detection process, through establishment of skin
of human face in color images
model and segmentation of skin region.

A MATLAB based face Jawad Nagi, Syed Khaleel It is a image based approach, using artificial
5 recognition system using image Ahmed and Farrukh Nagi intelligence by removing redundant data from face
processing and neural networks. through image compressing using 2D-DCT.

Face detection in color images Rein-Lien Hsu, Mohamed

It uses a skin tone color model and facial feature to
6 Abdel-Mottaleb and Anil
detect color images using face detection algorithm.
K.Jain

46
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 03 Issue: 02 December 2014, Page No. 46-48
ISSN: 2278-2419
III. PROBLEM WITH HUMAN SKIN In image detection process a 2D histogram shows the
DETECTION relationship of intensities at the exact position between two
images. The 2D histogram [4] is commonly used to compare
1. A reliable human skin detection method that is adaptable to
two channels in a multi-channel images, where the x-axis
different human skin colors and illumination conditions is
represent the intensities of the first channel and the y-axis the
essential for better human skin segmentation.
intensities of the second channel. As a comparison, a 1D
2. The existing methods require high computational cost. We histogram is nothing more than counting how many voxels
discuss a novel human skin detection approach that combines a with a particular intensity occur in the image. The intensity
smoothed 2-D histogram and Gaussian model, for automatic range of the image is divided in bins. A voxel then belongs to
human skin detection in color images. the bin if its intensity is included within the range the bin
represents. The 2D histogram is the same as the 1D histogram
3. In our approach, an eye detector is used to refine the skin
with the difference that it counts the occurrence
model for a specific person. The proposed approach reduces
of combinations of intensities. To compute a 2D histogram the
computational costs as no training is required, and it improves
images need to be equal in size.
the accuracy of skin detection despite wide variation in
ethnicity and illumination. B. Gaussian Approach
4. The first method to employ fusion strategy for this purpose. This is a two-dimensional version of the Gaussian
Qualitative and quantitative results on three standard public function[5]. All that really matters here is the general image. It
datasets and a comparison with state-of-the-art methods have peaks in the centre and decays as you move away. You can
shown the effectiveness and robustness of the proposed control how quickly it decays by varying the σ parameter.
approach.
A Gaussian function is a function of the form:
A. Type of problem
Low Accuracy: False skin detection is a common problem
when there are a wide variety of skin colors across different
ethnicity, complex backgrounds and high illumination in im-
age(s).
Luminance-invariant space: Some robustness may be For arbitrary real constants a, b, c, d.
achieved via the use of luminance invariant color space, how- The graph of a Gaussian is a characteristic symmetric "bell
ever, such an approach can withstand only changes that skin- curve" shape that quickly falls off towards zero. The para-
color distribution undergoes within a narrow set of conditions meter a is the height of the curve's peak, b is the position of the
and also degrades the performance. center of the peak, and c  controls the width of the "bell".Gaus-
Require large training sample: In order to define threshold sian functions are widely used in statistics where they describe
value(s) for detecting human skin, most of the state-of-the-art the normal distributions, in signal processing where they serve
work requires a training stage. One must understand that there to define Gaussian filters, in image processing where two-di-
are tradeoffs between the size of the training set and classifier mensional Gaussians are used for Gaussian blurs, and in math-
performance. ematics where they are used to solve heat equations and diffu-
sion equations and to define the Weierstrass transformUsing
IV. METHODOLOGY Gaussian processes for noisy 2-dimensional Interpolation Re-
Firstly we will upload an image into the software and then we call the decomposition of the observations into spatial and in-
have to get the statistic of an image by the use of preprocessing dependent components we are striving for yi = si + ni.
(RGB mean value) the selected image. After getting the mean In order for Gaussian processes to be used efficiently we
value, we can it for detecting the human skin image. The 2D assume fnigNi=1 are Gaussian in addition to being
histogram and Gaussian approach is implemented for detecting independent, identically distributed with zero mean. We will
the human skin. see that the variance of ni will be estimated from the data
automatically. Initially, we will also assume si = f(xi) for some
unknown (latent) function of the two coordinate inputs x;
hence, ni will denote the departures of yi from f(xi) and our
task can be seen as determining the value of f at all data point
locations, what on may call the two-dimensional noisy
interpolation problem. The approach of parametric methods is
to parameterize a class of functions conjectured to contain the
right f and use the data to estimate its parameters. Coming up
with the right parametric form is generally difficult. By
contrast, Gaussian process regression discovers f without
explicitly stating a parametric form but rather by vaguely
specifying its expected behavior through a “distribution over
functions”.
Fig 1: System flow V. RESULT AND DISCUSSION
A. .2D Histograms
47
Integrated Intelligent Research (IIR) International Journal of Data Mining Techniques and Applications
Volume: 03 Issue: 02 December 2014, Page No. 46-48
ISSN: 2278-2419
Giving the input by selecting the image for testing the expected work. By adding the layout, the result is become clearer and
result, this is implemented for the discussion. detail for the user.
Step 1: REFERENCES
[1] Ahmed Elgammal, Crystal Muang and Dunxu Hu Department of Com-
For implementation first we have taken an original image to puter Science, Rutgers University, Piscataway, NJ, 08902, USA Interna-
gather RGB mean value in the preprocessing section. tional Journal of Computer Vision (IJCV) 46(1) (2002) 81–96
[2] Fleck, M.M., Forsyth, D.A., Bregler, C.: Finding naked people. In: Pro-
ceedings of the
[3] European Conference on Computer Vision (ECCV). (1996) 593–602 Ab-
del-Mottaleb, M., Elgammal, A.: Face detection in complex environ-
ments from color lmages. In: Proceedings of the International Conference
on Image Processing (ICIP). (1999) 622–626
[4] Sriram Jayaram, Stephen Schmugge, Effect of Colorspace Transforma-
tion, the Illuminance Component, and Color Modeling on Skin Detection.
Proceedings of the 2004 IEEE Computer Society Conference on Com-
puter Vision and Pattern Recognition (CVPR’04).
Fig 2: Original image [5] Lee, Y., Yoo, S.I.: An elliptical boundary model for skin color detection.
In: Proc.of the Int. Conf. on Imaging
[6] Science, Systems, and Technology. (2002).

Fig 3: Mean Values(RGB)

Step 2:
After getting the RGB mean value, we are being proceeded the
2D Histogram for the original image.

Fig 4: 2D Histogram
Histogram (Fig 4) is the 2D Histogram image which is taken
from the original image (Fig 2). In the 2D Histogram, the front
image and the back image is been separated from the original
image.
Step 3:
We can separate the Human skin successfully using Gaussian
Approach (Fig 5) from the Histogram(Fig 4).

Fig 5: Human Skin Detection

VI. CONCLUSIONS
As a conclusion this project is almost fully completed success-
fully. The layout which can be added by developed the 2D his-
togram(fig 4) of Red, Green, and Blue (RGB) colors and
Gaussian approach in other to calculate the surface area of the
input image and displays that values in both decimal and hexa-
decimal to the final result also recommended as the future

48