CHAPTER 1

INTRODUCTION
1.1. OVERVIEW
In past few years, identity theft and terrorism like activity have been increased,
therefore need of biometrics based security system has been a one of different areas of
research. The Security system biometrics is a modernizer technology which provides
the automatic system to distinguish between a genuine person and a fraud. There are
two parts in automated face recognition system, first one is face detection and another
one is recognition.
In this thesis work we have used face detection system to image accessing. To
identify a face from using a different online regulation system or a offline image the
important factor is prediction of the skin area. There are different technique which
discussed in different chapters, but the key factor is skin color.

Therefore, this thesis presents a skin based segmentation algorithm for face
detection in color images with detection of multiple faces and skin color regions. It
has been already proved that skin color to be a useful and robust cue for tracking,
localization and face detection. In Face recognition the given input image is checked
whether it contains human face or not, if so, give the result as location of the human
face.

Basically due to lots of research, wide variety of applications (like in

computers, security ,communication, law enforcement, smart home application and ,
image database management ,automatic control systems etc) and difficulties in face
detection have become interesting topics of research for the researchers in past
decade.

There are many places which are using face detection especially the websites
hosting images like Picasa, photo bucket and Face book. Actually Face Detection is
first essential step in Face Recognition system for localizing and extracting the
features of given image.

There are lots of discrepancy in appearance based approach but the face
detection approach is not a simple and easy. It consists of a different factor such as
pose variation, facial size, image orientation, structural components, illuminating

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

1
conditions, occlusion, facial expression found in the image, scene and complexity of
images background and others.

1.1.1 Objective
A lot of work has been done related to this work, but the process which has
been adopted earlier having some drawbacks like accuracy & speed of the cropping of
the faces. In this work highlighted to overcome the earlier failures. This work focused
to increase the accuracy & speed of the cropping of the faces.

1.1.2 Motivation
Now days Biometric face recognition is one of the most recent and strapping
application for skin and face recognition. This applications used for pattern matching
of images. A wide-ranging description of biometric acknowledgment based on the
most popular human biometric traits is provided. A huge prominence was placed on
the face uniqueness, predominantly on the state of the skill of the component-based
face recognition, while the problem of being acquainted with faces under gross
variations remains largely unsolved. A demonstration system as proof of concept that
such systems are now becoming practical have been developed. A system capable of
reliable recognition, with reduced constraints in regards to the facial position and
expression of the face with the illumination variation and background of the image
have been implemented.

Proposed work focused on implemented work and compares existing face

recognition algorithms which are widely used and subject of interest. Some
applications of automatic facial feature detection are as follows:-

Computer generated facial expressions can be generated by tracking features on the

human face and then replacing the facial texture with a similar face.

The location of key feature points on the face could be used to aid the recognition of
human facial expressions.

Algorithms which identify and crops the human beings faces.

1.2 Types of Face Detection

Face detection methods classified on various approaches for detection which

are generally classified into four categories:

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

2
1.2.1 Feature Based Methods

1.2.2 Knowledge Based Methods

1.2.3 Machine Learning Methods

1.2.4 Template Matching Method

Figure 1.1 shows the classification of face detection methods. This method is
based on Feature-based techniques, which focus on the combination of edge,
integration of skin color, symmetry measures, motion of faces with degree, space gray
level dependence matrix (SGLD), feature analysis and point distribution models
(PDMs).

Figure 1.1: Face Detection Methods

The second method is Knowledge based which includes multi-resolution

based method as shown in figure 1.1. The problem of face detection in still images is
more easy when compared to the problem of face detection in video or real time
applications since motion information can lead to probable regions where a face could
not be located easily.

The third method is based on Appearance-based Techniques which includes

neural networks, linear subspace methods like Eigen faces, fisher faces, Eigen faces
with background learning and statistical approaches like support vector machines,
higher order statistics, distribution based and information theoretical approaches. The

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

3
fourth method is based on Template matching, which use predefined templates such
as Shape templates, Deformable template like Active shape model etc.

1.2.1 Feature Based methods

Basically feature based method is most widely used method because it extracts
the necessary facial features, and their mutual distances using statistical model are
calculated, which is not affected by any type of variation like lighting condition,
vision ,cause, nature and other factors. All these method categorize according to the
extracting features of images. With the help of Feature based technique, the features
of face to gain the required knowledge of face are extracted. Features may be skin
color, face shape, and expression or facial features like eyes, nose, eyebrow and lips
etc. Feature based methods are used for real time systems.

1.2.1.1 Low Level Feature Analysis

1.2.1.1.1 Edges

As a useful primitive feature in computer vision, edge representation was

applied to early face detection system. Later, this work was upgraded and a
hierarchical frame work proposed to trace the human head line. This approach
includes a line follower which is implemented with a curvature constraint. Some more
recent examples of edge-based techniques can be found. Edge detection is an
important step in edge-based techniques.

1.2.1.1.2 Skin Color

Human skin color can be used as an effective feature for face detection, and
related applications. Human faces are colored depending upon the region of their
habitation, but different studies show that the major difference exists in the intensity
rather than the chrominance.

1.2.1.1.3 Motion

Motion information is a convenient way of locating moving objects when a

video sequence is provided. Using this information the face searching area can be
narrowed. The simplest technique to achieve motion information is frame difference
analysis. M. Turk and B. Sankur employed, improved frame difference analysis for
face detection.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

4
 Besides face region, also employ frame difference to locate facial features,
such as eyes. Another way of measuring visual motion is through the estimation of
moving image contours. Compared to frame difference, results generated from
moving contours are always more reliable, especially when the motion is
insignificant.

1.2.2 Knowledge Based

Knowledge based technique depends on how we are using rules for extracting
the features of human face. With the help of simple rules of these extracting features,
we can easily make relationships in features of face. The image of Facial appearance
as given input image as extracted and face of the given candidates images is identified
based on the given coded rules.

For image deduction the verification process is frequently applied to reduce

false detections. Translation of human knowledge into well-defined rules is still
causes some kind of difficulties. In detailed rules (i.e., strict), it may fail to detect
faces which do not pass all the rules but if the rules are moreover general then they
may give many false positives.

For example, in an image we have different features like eyes that are
symmetrical to each other, nose, lips, and mouth. These features of face represent a
relative distance, position and relationships respectively. After discover features from
the face image, verification is done to condense false recognition. The knowledge
based approach is superior for frontal image and not varying image or lighting
condition.

1.2.3 Image or Appearance based Method

In Image based approach the predefined standard face pattern and database
which is used for match with the section in an different image to determine whether it
is a actual face or not. There are many Training algorithms available which categorize
different regions into face and the different non-face classes. This method depend on
multi-resolution window scanning for the detection of faces, so these techniques more
efficient in detecting faces but execution is slow as compare to feature-based
techniques.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

5
 Eigen-faces, Distribution based method, Hidden markov model (HMM) and
neural networks etc are examples of image-based techniques. Hidden markov model
(HMM) approach is simple to implement, and also good for frontal view but it cannot
well compact with variation in shape scale and pose.

1.2.4 Template matching method

Template matching methods finds the similarity between the input images and
the template images (training images). Template matching method uses the
correlation between the input images and stored standard patterns of whole face
features to determine the presence of whole face features. Predefined templates as
well as deformable templates can be used.

This method can be used for both face detection and face locations. In this
method, a standard face (such as frontal view) can be used. The advantages of this
method are that it is very simple to implement the algorithm, and it is easy to
determine the face locations such as nose, eyes, eyebrow and mouth etc based on the
correlation values.

It can be applied on the various variations of the images such as pose, scale,
and shape. Sub-templates, multi resolutions, and multi-scales have been proposed to
achieve the shape and the scale invariance. Smita et.al, offered the idea of a
localization method which is based on a shape pattern of a front view face. We can
use different methods of edge detection to find out the edges of face like Sobel filter is
used to extract the edges.

1.3 Color space model and Color transformation

The important technique for face detection is Skin color detection in different
color images. There are so many techniques that focus on locating skin color regions
for the given input images. These techniques use HSV or YIQ formats and which the
concept of different color combination and different component with like input color
image generally user RGB format. If the lighting condition changes the face detection
may be fail because it is depend upon RGB component.
Basically, color pattern is not easily perceived and interpreted by human brain
because it is a physiological phenomenon that is not fully understood by brain. The
physical nature of color can be understood on the basis of practical and theoretical

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

6
results. In our eye, we have cone sensor which is responsible for color division. This
is divided into three principle sensing categories red, green, and blue (RGB).

Our cone sensors are 65% sensitive to red light, 33% to green light and only
2% for blue respectively. Due to these characteristics of the light for human eye,
colors are seen as variable combination of three primary colors red(R), green (G), and
blue (B). The main function of RGB colour model is for the scarted image
representation, sensing the image and display the image in different electronics media
like PC (personal Computer) ,Television, Face book, Twisted .

Main purpose of the RGB color model is for the sensing, representation, and
display of images in electronics system such as television, computers and also in
photography.

Now when we are working with color images, we need to describe the
information of each pixel of that color image. When human eyes perceive the color, it
is seen in 3-D due to those three-color components red (R), green (G), and blue (B).
We have three characteristics; Hue (H), Saturation (S), and brightness or value (V)
which is used to distinguish one color from another color. Nowadays, a different type
of color model is used which is widely classified and have broad applications areas.

We can represent three components of color in several ways like RGB color
model, HSV/HSI color model, YUV color model, TSL color model, CMY color
model, YIQ color model and YCbCr model etc

1.3.1 The YCbCr Color Model

YCbCr color model wedge the digital image into two parts, luma or luminance
component and chroma or chrominance component. Due to luminance and
chrominance effect, it becomes most important in digital video camera to handle
video information. In YCbCr color model, Y stands for luminance effect, Cb and Cr
stands for chrominance effect or blue-difference and red-difference respectively.
Although Y luma represent black and white color, while Cb, and Cr represents RGB
colors as shown in figure 3.7.
This thesis used the color space YCbCr components this color spaces save
computation time because it is using one of the existing Matlab functions. Here Y
represent luminance information, Cb and Cr are chrominance information for the

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

7
YCbCr color space. Finally the RGB components convert to YCbCr components and
luminance information can be easily de-embedded.

Y = 0.29R + 0.587G + 0.114B

Cb = -0.16R - 0.332G + 0.500B
Cr = 0.50R - 0.419G - 0.081B

Figure 1.2: YCbCr Color Model

For YCbCr Color Model, the skin color recognition procedure each pixel
divided as two class which is based on skin color combination and different. The skin
color resolute by using the standard divergence of Cb and Cr component, obtained
using 164 training faces in 7 input images and skin colour for the recognition window
determined based on Cb and Cr components of 164 faces are plotted in the color
space and means. The given Fig.1.2; focus on histogram distribution and Fig. 1.4.,
Fig. 1.3 focus on Skin pixels in YCbCr color space.

Figure 1.3: Skin pixel in YCbCr Color Space

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

8
 Figure 1.4: (a) Figure shows Histogram distribution of Cb.

Figure 1.4: (b) Figure shows Histogram distribution of Cr.

Figure 1.4 focuses on the training image and its result for color segments and
Fig.1.5 focus some non-skin objects that predictably distribute the result as their color
fall and also highlighted on skin color space.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

9
 Figure 1.5: Color Segmentation Result Of A Training Image.

1.4 Image Segmentation

The next step is to separate the image blobs in the color filtered binary image
into individual regions.

Figure 1.6: Small Regions Eliminated Image.

The process consists of three steps. The first step is to fill up black isolated
holes and to remove white isolated regions which are smaller than the minimum face
area in training images.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

10
 CHAPTER 2
LITERATURE REVIEW

Sung and Poggio [1] proposed and successfully implemented Gaussian

clusters to model the distribution of facial and non face patterns, that helped on
differentiating between real and non-real faces. Rowley et.al [2] used artificial neural
network for face detection. The face detection procedure got improvised by having a
mesh kind of framework to compare with the specified data or image available. Yang
et.al [3] provide the idea of confidential face recognition methods in four categories.
(i) Knowledge based (ii) feature invariant (iii) template matching (iv) appearance
based. Ming-Hsuan Yang et.al, conducted a survey of Detecting faces in images and
also compared different techniques like Face detection, face recognition, object
recognition, view-based recognition, statistical pattern recognition, machine learning
etc [3]. Lu et.al [4] used parallel neural network for face recognition. With all the
contribution from the field of neural network, introducing parallel neural network was
another step near to getting precise results for recognizing faces and accurate outputs.

The methodology Zhao et.al [5] proposed Linear Discriminant Analysis

(LDA) for face recognition. H C Vijay Lakshmi et.al, proposed a idea Edge Detection
Techniques with using the Segmentation Algorithm, this paper used the concept of
Multiple Face Detection in Color Images and Skin Tone Regions using Color Space
Segmentation Algorithm for Multiple Face Detection in Color Images with Skin Tone
Regions using Color Spaces and Edge Detection Techniques, in which different color
space models, specifically, HSI and YCbCr along with Canny and Prewitt edge
detection techniques are used for better face detection [6]. Iyad Aldasouqi et.al,
presented a Smart Human Face Detection System, in which they explained Digital
image processing and fast detecting algorithms based on different colour model which
used HSV Color model without admit defeat the speed of face recognition [7]. The
comparative analysis is a good read to know all the techniques available for face
related recognition and detection and also, other procedures applied in these area.
With field being recognized and applied in various fields, the improvements are
beneficial for the end users.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

11
 Diedrick Marius et.al, worked on Face Detection Using Color Thresholding,
and Eigen-image Template Matching, compared and used the YCbCr, HSV color
model for face detection and segment an image based on Skin color [8]. These are the
most used technique in skin color detection and segmentation which is an important
part of the project. Arti Khaparde et.al, proposed an algorithm based on color
segmentation and morphological operation like closing, opening and connecting etc
and applied segmentation on HIS and YCbCr color model [9].

The method too is combined methodology of two procedures and the output
achieved is quite effective and efficient in gaining the desired results. Amol
Dabholkar et.al, worked on Automatic Human Face Detection and Tracking, in which
the features using motion based and model-based algorithms were extracted and also
used Kalman filter for estimation of the feature motion [10]. Aamer S. S. Mohamed
et.al, presented a idea of neural networks and Discrete Cosine Transfer (DCT) which
uses two dimensional skin based data extraction by using the Cb and Cr feature
vectors and DCT coefficient [11].

Smita Tripathi et.al, proposed a Face Detection using Combined Skin Color
Detector and Template Matching Method in which, they discussed Skin Detection
with the help of YCbCr color model and edge detection techniques for face detection
[12]. Rachid Aliradi, Naima Bouzera, Dr Abdelkrim Meziane[13], Chaoyang Zhang,
Zhaoxian Zhou, Hua Sun, and Fan Dong [14], Face distinguishment has gained a
great deal of consideration in biometrics and workstation vision.

A considerable measure of face distinguishment calculations have been

produced throughout the previous decades. This paper surveys three established
routines Main Part Investigation (PCA), Straight Segregate Dissection (LDA), and
Versatile Bundle Chart Matching (EBGM). Three calculations are actualized with
MATLAB.

Chen Da-jin/Chen Si-yu/Su Yun-huan, Peng Min-jing [15], so as to take care

of the issue that Omni-directional confronts, which was in pictures with complex
connection, couldn't be discovered, an eye-center based face discovery model was
proposed. In the proposed model, the strategy of HSI based skin discovery joined
together with eye-center identification was utilized to identify eyes, and afterward
picture turn, characteristics extraction from pictures and neural system based

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

12
characterization were connected to execute face recognition and face affirmation, and
acquired the objective that Omni-directional countenances could be discovered.
Finally, an analysis was led to check the proposed model with MATLAB 2011ra, and
the come about that location correctness was 95% demonstrated that the model was
successful.The objective is to recognize human appearances from avatar faces. Our
methodology could attain high correctness, 99%, utilizing short of what 1% of the
pixels in each one picture.

This was gotten in the wake Salem Alelyani, Huan Liu [16], Troupe
characteristic determination is known for its heartiness and generalization of
exceedingly faultless prescient models. In this paper, we utilize distinctive lter-based
peculiarity choice systems in a group way to enhance face distinguishment. of
uprooting immaterial gimmicks which is known to corrupt taking in execution and
model dependability.

Emir Kremic, Abdulhamit Subasi, and Kemal Hajdarevic [17], The point of
this paper is to present and propose customer server model and to contrast it and the
latest customer server models for face distinguishment with a GPG foundation
which utilizes security private key (symmetric encryption) with fundamental reason to
safely transmit picture (client face) over the system. Besides in the face
distinguishment calculation is actualized Standard Segment Dissection (PCA)
calculation for face distinguishment. Proposed framework has been tried on the cell
telephone with Android OS stage, utilizing past examination encounters where
framework was at first produced for DROID emulator. The execution of the PCA is
carried out on the MATLAB.

R. Kjedsen and J. Kender accomplished the idea of a review on Pixel-Based

Skin Color recognition Techniques, in which they concluded that skin color, is very
useful and cue for face detection, locating and tracking and also compared the
different methods and techniques of face detection [18]. K. Sobottka and I. Pitas,
described the Face Detection method with Skin Color Information and also
implemented Adaptive boost algorithms and six segmented rectangular filters for
better performance [19]. H. Wang and S.F Chang, compared the different color
models like HSV, YCbCr, and HIS, combined these models and developed a new
Robust Skin Color Based Face Detection algorithm which gives improved accuracy

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

13
compared to other techniques [20]. Q. Chen et.al, presented the concept of Color
Based Segmentation automatic Face Detection Using and Template/Energy
Thresholding and 28 they compared the RGB, HSV and YCbCr Color model
techniques for face detection and also applied some morphological operation for
better accuracy [21]. S. Kawato and J. Ohya, presented the concept of Edge
information system and Cluttered Color Images on Human Face Detection in
Cluttered, in which they developed HSV color model based face detection using edge
detection technique [22].

The methodology is quite simple yet effective to have distinguished output of

individual features from a group of multicolor images or a image with more number
of persons and a rainbow of colors. A. Albiol et.al, proposed a an algorithm in which
they could detect the multiple faces in color images with various illumination and
discussed triangle based segmentation and multilayer neural network using HSV
Color space [23]. The method is great in getting all the details related to an individual
from a group colored image. J. Yang and A. Waibel, compared and combined the
Traditional Method of HSV Histogram Equalization with provide the concept of
Contact Image retrieval with using Adaptive HSV segmentation and Kekre Transform
for [24]. The paper gave a thorough view of the two methods with an elaborate
methodology and how different methods can be combined to get a desired result. M.
J. T. Reinders et.al, presented a concept of Bayesian Theorem using different color
Image Segmentation and probability density function based (KDE) Kernel Density
Estimation in which they used probability density function based on Bayes theorem
for color segmentation and also used HSL color model for face detection [25].

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

14
 CHAPTER 3
PROBLEM IDENTIFICATION

The challenges associated with face detection can be attributed to the

following factors:

i. Pose- Actually face has various position which vary due to the camera
positions or lower resolution camera and face pose (like frontal view, 45 degree,
profile, upside, own), due to various position of camera, Nose and an eye may be
some partly or wholly affected.

ii. Facial Presence or absence of Facial components (Structural

components) Some structural features with Facial features such as beards,
moustaches, growing hair and glasses might or possibly will not be present and there
is a great deal of inconsistency among these apparatus including nature, shade, and
size, color, height and resolution of images .

iii. Facial expression- Some internal problems of human directly affect the
human expression or facial expression.

iv. Scattering - Different faces may be partly scattered by other objects. The
group of people may be cause of scattered property of faces.

v. Image orientation- We have different types of camera at different places

but the rotation and optical axis of that camera also differ so face images may vary
due to rotation and optical axis.

vi. Imaging conditions- Some factors should be kept in mind at the time of
capturing an image that affected the lighting of images due to some region (like
range, source allocation and power of camera) and some other camera uniqueness
property (like inbuilt sensor in camera , capturing response time of images , lenses
using for images) which affect the appearance of a face.

The system related to face processing like identification, recognition,

segmentation etc. use some training algorithm which are used for template matting of
image pattern and increase the performance of the system. Generally it uses two
approaches concurrently for solving the problem, which are :-

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

15
 1. Pattern and Template Classification: For Correct pattern and template
classification need different classification techniques those recognize identify and
manage individuals which are in the database and manage image in database.

2. Reject False Scattering Image: Delete unusable images of individuals

which are not part of the database. This concept reduces effectiveness of a system by
increasing performance during run time which is a big drawback.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

16
 CHAPTER 4
METHODOLOGY

4.1 Overview of flowchart

Biometrics based security system is an example of visual pattern recognition
task. The consideration of three-dimensional human face depends upon varying
clarification, different pose of images and different expression. All these face
reorganization depend upon the range and selection of input data sources.
When taking at some photometric information of images reorganization like
face which analysis the subspace of images and reside in manifold of the facial
reorganization of Image. For skin detection generally face recognition algorithm is
used which is frequent use now days. The projected skin recognition based training
face recognition algorithm is mentioned in fig. 4.1.

This figure highlighted working principal of partition of layers to detect skin

area. The median filtering applies when the skin image once detected and passed. This
figure focus consequences of this step are like binary image.

Read Image

Convert Image into YCBCR

Colourspace

Seperate Layers

Detection of Skin area

Applying Erode Corrosion

Apply Median Filtering

Remove Pixel Area less

than 1000 Pixel

Face Segmentation and Labelling

Extract and Crop Face

End

Figure 4.1: Flow Chart of Methodology

According the image face detection for biometric information system, the need
of small groups of 1000 pixels and remaining block should be removed, Each pixel is
label according the level of skin label face and the exerted the original feature of face

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

17
then calculate the index value of original image . To perform the face detection and
cropping, following steps are performed:

The above explained face detection flowchart is implemented using

MATLAB.

4.2 ALGORITHM

The consequent sequence of steps as focus below in Fig. 4.2. The main steps
are as follows:

a. Read Image
b. Convert RGB image into YCBCR color space for better skin area
detection
c. Divide the layer different layer separately like layer Y, layer CB and layer
CR.
d. Take the value of CR and CB layer are better skin properties.
e. Assign skin parameters of intensities is 0 0r 1.
f. Assign remaining pixel of the pixels 0 or null.
g. Removing pixels are 2x2 then apply Erode corrosion.
h. Discard the value of which has 1000 pixels or less.
i. Calculate the pixel value of rest of group which has same label.
j. Find out the original pixel for calculating the pixel value.
k. The extracted pixels are face k.

Figure 4.2: Algorithm of Face Segmentation

First of all image is taken from different sources, one of the good property of
this approach is that it can work on images taken from any camera quality. After that
Method is applied for discarded the pixel value of 2x2 Convert RGB image into
YCBCR color space for better skin area detection. Separate the layer Y, CB and CR
because CB and CR are having better skin properties.

The RGB and YCbCr color space is good for input transform and give
accurate result but it is necessary RGB color space converted into YCbCr color space
and index value depend upon the input color detection which take the component
range [0.0- 255.0] then calculate the transforms value YCbCr color space which is
allocated on of it into Y[0.0, 255.0], Cb[128.0, 127.0],and Cr[-128.0, 127.0]

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

18
respectively. The level-shifted down for Y-component is 128 and this can depend
upon the value range into the [-128.0, 127.0] range. The DWT generally used the
symmetric YCbCr color space which is assign input level-shifted depend on the next
predication of DWT. The given matrix equation is shown in the following figure.

Figure 4.3: Matrix equation for RGB to YCbCr

4.3 EROSION OPERATION

The Erosion process is similar to dilation process. Erosion shrinks or thins a
binary image or object. As in dilation, the manner and extent of shrinking is
controlled by the structuring element. The output of erosion has a value 1 at each
location of the origin of the structuring element, such that the structuring element
overlaps only 1-valued pixels of the input image.
4.4 MEDIAN FILTERING
Median filtering is applied to remove noise from the images. The median
filter is a nonlinear filtering technique, often used to remove noise. The image noise
reduction improves the color of image processing system and specially face
reorganization system before pre-processing step. Now days the Median filtering
concept is using for digital image processing because, under certain conditions, it
preserves edges while removing noise (but see discussion below).
Assign the skin color intensities value for pixels 0 or null. Remove 2x2 pixels
then apply Erode corrosion. Discard the value which has 1000 pixels or less.
Calculate the pixel value of rest of the group which has same label. Find out the
original pixel for calculating the pixel value.

Skin Color Segmentation using YCbCr is the next step applied on image
obtained from previous step. A skin color classification algorithm with color statistics
gathered from YCbCr color space. The skin value of face detection with the same
index value and belonging to skin region with similar Cb and Cr values. Therefore it
can be depend upon the value of be seen that skin color model based on the Cb and Cr
values provides good coverage of different human faces.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

19
 The thresholds values of face reorganization have been chosen as [Cr1, Cr2]
and [Cb1, Cb2]. The result threshold values [Cr, Cb] fall within the previous taken
thresholds values. For the skin face reorganization it is better to take color image. The
constraint of this algorithm is that the image should contain only face as the skin
region.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

20
 CHAPTER 5
RESULT AND PERFORMANCE ANALYSIS

In first step input image is taken as shown in figure 5.1

Figure 5.1: Sample 1-Original Image

Then Different Color components are extracted from the original image

Red Component Green Component Blue Component

Figure 5.2: RGB component of Sample 1

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

21
 YCbCr color pattern is extracted from the RGB image as shown in figure 5.3.
RGB image is converted into YCBCR color space for better skin area detection.
Separate the layer Y, CB and CR because CB and CR are having better skin
properties.

Figure 5.3: YCbCr Color Component of Sample 1

After that binary image is obtained as shown in figure 5.4

Figure 5.4. Binary image of Sample 1

Finally, after applying face detection algorithm face is detected as shown in figure 5.5

Fig 5.5 Face Detected of Sample 1

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

22
 (a) (b) (c)

Fig 5.6: Combined Result of Sample 1 (a) Original Image (b) Skin Like Area (c)
Face Detected

(a) (b) (c)

Fig 5.7: Sample-2 (a) Original Image (b) Skin Like Area (c) Face Detected

(a) (b) (c)

Fig 5.8: Sample-3 (a) Original Image (b) Skin Like Area (c) Face Detected

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

23
This algorithm can be applied successfully in multiple faces also as shown in figure
5.8.

Fig 5.9. (a) Sample 4-Original Image

Fig 5.9. (b) Sample 4-Skin Like Area of all Images

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

24
 Fig 5.9. (c) Sample 4-Face Detected

Fig 5.10. (a) Sample5-Original Image

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

25
 Fig 5.10 (b) Sample 5-Skin Like Area

Fig 5.10 (c) Sample 5-Face Detected

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

26
 Fig 5.11 (a) Sample 6-Original Image

Fig 5.11 (b) Sample 6- Skin Like Area

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

27
 Fig 5.11 (c)Sample 6- Face Detected
To evaluate the performance of the proposed algorithm, following parameters
are taken in to consideration. The parameters under consideration are number of faces,
detected faces, number of repeat faces, false positive, and time to execute the
algorithm and accuracy of face detection.

Table 5.1 Performance Evaluation of the Algorithm

Number of Faces 06
Detected Faces 06
Number of Repeat Face 00
Fig 5.9
False Positive (Wrong Detection) 00
Time to Execute 1.3298
Accuracy 100%
Number of Faces 09
Detected Faces 06
Number of Repeat Face 00
Fig 5.10
False Positive (Wrong Detection) 02
Time to Execute 1.2807
Accuracy 75%
Number of Faces 06
Detected Faces 05
Number of Repeat Face 00
Fig 5.11
False Positive (Wrong Detection) 00
Time to Execute 1.1213
Accuracy 83%

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

28
Table 5.1 shows the performance evaluation of the algorithm and the overall accuracy
of the algorithm is found out to be 73.68%.

Figure 5.12: Performance Evaluation Graph

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

29
 CHAPTER 6
CONCLUSION

From the above results it is clear that the proposed methodology is

completely free from training database, means it can be applicable in various systems
where face area need to be extracted out. Thus, no particular database is required with
any specified size or property.

The technique having advantage of extraction of multiple faces in a given

image and so, a lot of work becomes easy as getting a result doesnt depend on
finding an image with a single face.

The proposed algorithm is high in accuracy and takes very less time for
execution. Table 5.1 shows the performance evaluation of the algorithm and the
overall accuracy of the algorithm is 73.68 % for multiple images and 99.11% for
single images. An important feature of the technique is that it avoids the images
which are not clear or overlapped with another skin area. It also doesnt mix two or
more skin color to give as an output.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

30
 CHAPTER 7
SCOPE AND FUTURE WORK

SCOPE
The algorithm is very useful in detecting faces which can be used in
various fields such as security monitoring, identification in shaded or low light areas,
etc. The training free part of the algorithm makes it unique in itself making it helpful
where the condition is to take less time to recognize a person or to find a particular
face on demand.
The algorithm overall is easy to implement and the use is very handy
making it user-friendly hence, the areas with shortage of technical skills can have a
great use of the image identification without much prior training or development of
the resources.
The algorithms used such as YCbCr and Median Filtering used here are
well synchronized such that the output of each can be monitorised and visualized to
get the result accordingly or to find a fault wherever required.

FUTURE WORK
In future the advancement in the skin detection algorithm with the
integration of other feature of eyes, nose, mouth and chin makes this technique more
accurate and robust with high speed detection and recognition of faces in a given
image with multiple face detection technology in a single attempt.

The technique is sometimes segmenting a single section of skin due to

difference in skin color area, which is a drawback if taken in account which can be
solved by improving the skin detecting mechanism and extraction process.
The system avoids images which are partially visible to lessen the
complication of identification but has a drawback which can be improved by
identifying partial images which are almost identifiable that is if image is almost say
more than 60% visible.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

31
 LIST OF REFERENCES

[1]. K K. Sung, and T Poggio, Example-based learning for view-based human face
detection, IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 20, pp. 3951, Jan. 1998.

[2]. H Rowley, S Baluja, and T Kanade, Neural network-based face detection,

IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, pp.
2338, Jan. 1998.
[3]. Ming-Hsuan Yang, David J. Kriegman, and N Ahuja, Detecting faces in
images: A survey, IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 24 pp. 3458, Jan. 2002.
[4]. J Lu, X. Yuan, and T. Yahagi, A method of face recognition based on fuzzy c
means clustering and associated sub-NNs, IEEE Transactions on Neural
Networks, vol. 18, pp. 150160, Jan. 2007
[5]. H Zhao and P C Yuen, Incremental linear Discriminate analysis for face
recognition, IEEE Transactions on System Man and Cybernetics B, vol. 38,
pp. 210221, Feb. 2008.
[6]. H.C. Vijay Lakshmi and S. Patil Kulakarni, Segmentation Algorithm for
Multiple Face Detection in Color Images with Skin Tone Regions using Color
Spaces and Edge Detection Techniques, International Journal of Computer
Theory and Engineering, vol. 2, Aug. 2010.
[7]. Iyad Aldasouqi, and Mahmoud Hassan, Smart Human Face Detection
System, International Journal of Computers, vol. 5, March 2011.
[8]. Diedrick Marius, Sumita Pennathur, and Klint Rose, Face Detection Using
Color Thresholding, and Eigen image Template Matching, IJERMCA vol 3, pp
28-33, vol 3, March 2014.
[9]. Arti Khaparde, Sowmya Reddy. Y, Swetha Ravipudi, Face Detection Using
Color Based Segmentation and Morphological Processing: A Case Study, Int.
Symposium on Computer Egg. & Tech., March 2010.
[10]. Amol Dabholkar and Kin Choong Yow, Automatic Human Face Detection and
Tracking, IEEE T-PAMI, vol 4, pp.780785, 1997.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

32
[11]. Aamer S. S. Mohamed, Ying Weng, Stan S. Ipson, Jianmin Jiang, Face
Detection based on Skin Color in Image by Neural Networks, Int. Conf. on
Intelligent and Advanced Systems, vol 6, pp. 779-783, Nov 2007.
[12]. Smita Tripathi, Varsha Sharma, Sanjeev Sharma, Face Detection using
Combined Skin Color Detector and Template Matching Method, International
Journal of Computer Application, vol. 26, July 2011.
[13]. Rachid Aliradi, Naima Bouzera, Dr Abdelkrim Meziane,Detection of facial
components based on SVM classification and invariant feature, 2013
IEEE/WIC/ACM International Conferences on Web Intelligence (WI) and
Intelligent Agent Technology (IAT), vol 3, pp:30-36, sep 2013.
[14]. Chaoyang Zhang, Zhaoxian Zhou, Hua Sun, and Fan Dong, Comparison of
Three Face Recognition Algorithms 2012 International Conference on Systems
and Informatics (ICSAI 2012) IEEE, vol 4, Aug 12.
[15]. Chen Da-jin/Chen Si-yu/Su Yun-huan, Peng Min-jing, A Fast Detection Model
for Omni-directional Faces The 7th International Conference on Computer
Science & Education (ICCSE 2012) pp:767-770, July2012. Melbourne,
Australia IEEE.
[16]. Salem Alelyani, Huan Liu, Ensemble Feature Selection in Face Recognition
ICMLA 2012 Challenge 2012 11th International Conference on Machine
Learning and Applications, IEEE, pp:588-591, Dec 2012 .
[17]. Emir Kremic, Abdulhamit Subasi, and Kemal Hajdarevic, Face Recognition
Implementation for Client Server Mobile Application using PCA proceeding
of ITI 2012 34th International Conference on information technology interfaces,
pp:435-440, June 2012.
[18]. R. Kjedsen and J. Kender, Finding skin in color images. Proc. of the 2nd Int.
Conf. 54 on Automatic Face and Gesture Recognition, pp. 312317, July1996.
[19]. K. Sobottka and I. Pitas, Extraction of facial regions and features using color
and shape information. Proc. of Int. Conf. on Pattern Recognition. vol 3,
pp:421-425, Aug 1996.
[20]. H. Wang and S.F Chang, A highly efficient system for automatic face region
detection in mpeg video. IEEE Trans. on Circuits and Systems for Video
Technology, vol 7, pp. 615628, Aug 1997.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

33
[21]. Q. Chen, H. Wu, and M. Yachida, Face detection by fuzzy matching. Proc. Of
the 5th IEEE Conf. on Computer Vision, vol 5, pp: 591-596, jun 1995.
[22]. S. Kawato and J. Ohya, Real-time detection of nodding and head shaking by
directly detecting and tracking the between-eyes. Proceedings 4th IEEE
International Conference on Automatic Face and Gesture Recognition, vol 4,
pp: 40-45, Mar 2000.
[23]. A. Albiol, C. A. Bouman, and E. J. Delp, Face detection for pseudo-semantic
labeling in video databases. Proc. Int. Conf. on Image Processing, vol-3, 607-
611, oct1999.
[24]. J. Yang and A. Waibel, A real time face tracker. IEEE Proc. of the 3rd
Workshop on Applications of Computer Vision, vol-3,pp:142-147, Dec1996.
[25]. M. J. T. Reinders, P. J. L. van Beek, B. Sankur, and J. C. A. van der Lubbe,
Facial feature localization and adaptation of a generic face model for model-
based coding. ignal Processing: Image Communication, vol 7, pp. 5774,
March 1995.
[26]. T. Sakai, M. Nagao, and T. Kanade, Computer analysis and classification of
photographs of human faces. Proc. First USA-Japan Computer Conference. pp.
27, Dec 1972.
[27]. I. Craw, H. Ellis, and J. R. Lishman, Automatic extraction of face-feature,
Pattern Recognition Le, vol 158, pp. 183187, Feb 1987.
[28]. J. Wang and T. Tan, A new face detection method based on shape
information. Pattern Recog. Lett, vol-21, pp. 463471, June 2000.
[29]. A. L. Yuille, P. W. Hallinan, and D. S. Cohen, Feature extraction from faces
using deformable templates. Int. Jour. Computer Vision, vol 7, pp. 104-109,
Jun 1989.
[30]. J. Choi, S. Kim, and P. Rhee, Facial components segmentation for extracting
facial feature. Proc. 2nd Int. Conf. on Audio and Video-based Biometric
Person Authentication, vol 4, pp 60-72, Jun 1999.
[31]. R. Herpers, K.H. Lichtenauer and G. Sommer, Edge and key point detection in
facial regions. IEEE Proc. of 2nd Int. Conf. on Automatic Face and Gesture
Recognition, vol 2, pp. 212217, Dec 1996.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

34
[32]. T. S. Jebara and A. Pentland, Parameterized structure form motion for 3d
adaptive feedback tracking of faces. Proc. of IEEE Conf. on Computer
Vision and Pattern Recognition, vol-5, pp. 144150, Jun 1997.
[33]. S. Satoh, Y. Nakamura, and T. Kanade, Name-it: naming and detecting faces
in news videos. IEEE Multimedia, vol 6, pp. 2235, Mar 1999.
[34]. J. L. Crowley and F. Berard, Multi-model tracking of faces for video
communications. IEEE Proc. of Int. Conf. on Computer Vision and Pattern
Recognition, Jun 1997.
[35]. R. Kjedsen and J. Kender, Finding skin in color images. Proc. of the 2nd Int.
Conf. on Automatic Face and Gesture Recognition, vol 7, pp. 312317, Jun
1996.
[36]. S. Kim, N. Kim, S. C. Ahn, and H. Kim, Object oriented face detection using
range and color information. Proc. of the 3rd Int. Conf. on Automatic Face and
Gesture Recognition, vol 3, pp. 7681,Dec 1998.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

35
 PUBLICATIONS

1. Ghanshyam Sahu, Bhanupratap Singh and Pradeep Tripathi, A Comprehensive

Review on Face Recognition Techniques, International Journal of Emerging
Technology and Advanced Engineering (IJETAE), Jun 2014, Volume 4, Issue 6,
pp: 216-218.

2. Ghanshyam sahu, A Research Review on Face Recognition Techniques,

International Forum of Researchers Students and Academician (IFRSA),
Published in IIJC, 2014, Volume 4, Issue 4, pp: 753-756.

3. Ghanshyam Sahu and Amrish Tiwari, Training Free Face Recognition And
Cropping Using Skin Identification Algorithm, International Journal of Scientific
& Engineering Research (IJSER), 2015, Volume 6, Issue 3, pp: 1581-1583.

Department of Computer Science & Engineering, VITS, Satna (M.P.) India

36