Illumination Invariant Hand Gesture Classification against Complex Background using Combinational Features

Illumination Invariant Hand Gesture Classification against Complex
Background using Combinational Features.
Anjali Patil
Electronics Engineering
DKTE TEI, Ichalkaranji
Maharashtra, India
anjalirpatil@gmail.com
Dr. Mrs. S. Subbaraman
Professor, Electronics Department
Walchand College of Engineering, Sangli.
Maharashtra, India
s.subbaraman@gmail.com
Abstract: Hand gesture classification is popularly used in
wide applications like Human-Machine Interface, Virtual
Reality, Sign Language Recognition, Animations etc. The
classification accuracy of static gestures depends on the
technique used to extract the features as well as the classifier
used in the system. To achieve the invariance to illumination
against complex background, experimentation has been
carried out to generate a feature vector based on skin color
detection by fusing the Fourier descriptors of the image with
its geometrical features. Such feature vectors are then used in
Neural Network environment implementing Back
Propagation algorithm to classify the hand gestures. The set
of images for the hand gestures used in the proposed research
work are collected from the standard databases viz.
Sebastien Marcel Database, Cambridge Hand Gesture Data
set and NUS Hand Posture dataset. An average classification
accuracy of 95.25% has been observed which is on par with
that reported in the literature by the earlier researchers.
Index Terms: Back-propagation, Combinational Features,
Fourier Descriptor, Neural Network, Skin color, Static hand
gesture
I. INTRODUCTION
Hand gesture recognition plays an important role
in the areas covering the applications from virtual reality
to sign language recognition. The images captured for
hand gestures fall into two categories viz. glove based
images and non-glove based images. Hand gestures
recognition also is correspondingly classified as glove
based recognition and non-glove based i.e. vision based
recognition.
Fig.1 Steps involved in proposed Hand Gesture Classification
In glove based approach, users have to wear
cumbersome wires which may hinder the ease and
naturalness with which the user interacts with computers
or machines. The awkwardness in using gloves and other
devices can be overcome by using vision based systems
that means video based interactive systems. This technique
uses cameras and computer vision techniques to recognize
the gestures in a much simpler way. [1] [2]. Vision based
approaches are further classified as 3D model (which is
exact representation of shape but is computationally
expensive) and appearance based 2D model which is
projection of 3-D object onto 2-D plane and is economical
computationally. This paper focuses on appearance based
methods for recognition of hand postures.
As shown in Figure 1, after capturing the image
of hand gesture, segmentation is done based on the skin
color. In the skin color detection process the RGB color
model is first transformed to appropriate color space and a
skin classifier is used to find a pixel is skin pixel or non
skin pixel. Skin color is the low level features extraction
technique which is robust to scale, geometric
transformations, occlusions etc. By the skin classification
the region of interest is observed which then is used to find
the boundary of the hand. After extracting the hand
contour, the Fourier Descriptors (FDs) are extracted and
combined with the geometrical features. The feature
vectors, thus formed, are given to artificial neural network
used as a classifier to classify the hand gestures.
Skin Color
Detection
Fourier
Descriptor
Geometrical
Features
Combination
of features
Classification
using ANN
Image Input
Gesture
Output
International Journal of Computer Science and Information Security (IJCSIS),
Vol. 16, No. 3, March 2018
63 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

The detailed implementation is explained in the
successive sessions. The main objective of this paper is to
present the contribution of the work done in direction to
classify the hand gestures with the help of skin color
correctly from images captured under different
illumination conditions. The system which will be robust
against variation in illumination and hence can be called as
illumination invariant.
The rest of this paper is organized as follows:
Section 2 presents the literature review on illumination
normalization and skin color detection. Experimental work
is discussed in Section 3. Detailed results are presented in
Section 4 followed by conclusions and future scope in
Section 5.
II. RELATED WORK
Detection of skin color in an image is sensitive to
several factors such as illumination conditions, camera
characteristics, background, shadows, motions besides
person dependent characteristics such as gender, ethnicity,
makeup etc. A good skin color detector must be robust
against illumination variations and must be able to cope up
with the great variability of skin color between ethnic
groups. Another challenge in detecting human skin color is
the fact that the objects in the real world which may be in
the background of the image can have skin tone colors, for
example, leather, wood, skin-colored clothing, hairs etc.
The systems not taking care of this aspect may have false
detections. The purpose of the research work is to identify
and classify the hand gestures with this type of
uncontrolled environment.
Image is represented in different color spaces
including RGB, normalized RGB, HSV, YCbCr, YUV,
YIQ, etc. Color spaces efficiently separating the
chromaticity from the luminance components of color are
typically considered preferable (Luma-Chroma model).
This is due to the fact that by employing chromaticity-
dependent components of color only, some degree of
robustness to illumination changes can be achieved.
Different Skin color models with comparison of their
performance have been presented by Terrillon et.al. in [3].
The detection and segmentation of skin pixels
using HSV and YCbCr color space has been explained by
Khamar et. al. in [4] wherein an approach to discriminate
color and intensity information under uneven illumination
conditions is highlighted. The threshold based on
histograms of the Hue, Saturation and Value (HSV) has
been to classify the pixels into skin or non-skin category.
The typical values of threshold applied to the chrominance
components followed the limits as 150 <Cr<200 &&
100<Cb<150. Chromacity clustering using k means of
YCbCr color space to segment the hand against the uneven
illumination and complex background has been
implemented in [5] by Zhang Qiu et.al . The different
experiments performed on the Jochen Triesch Static Hand
Posture Database II were reported with comparison in
terms of time consumed.
Bahare Jalilian et.al. detected regions of face and
hands in complex background and non-uniform
illumination in [6]. The steps involved in their approach
were skin color detection based on YCbCr color space,
application of single Gaussian model followed by Bayes
rule and morphological operations. Recognition accuracy
for images with complex background reported was 95%.
YCbCr color space was used in [7] by Hsiang et.al. to
detect hand contour based on skin color against the
complex background. Convex hull was calculated and the
angle between finger spacing and the finger tip positions
were derived to classify the hand gesture. The accuracy of
the recognition rate reported was more than 95.1%.
HSV based skin color detection was implemented
by Nasser Dardas et.al in [8], The method has been
reported to have real time performance and is robust
against rotations, scaling and lighting conditions.
Additionally it can tolerate occlusion well. The
thresholding proposed was H between 0o
to 20o
and S
between 75 and 190. The segmenting resulted in giving
the hand contour which was subsequently compared with
the templates of the contours of the hand postures. Four
gestures were tested by the authors which indicated an
average accuracy of more than 90%.
HSV based hand skin color segmentation was
used by Zhi-hua et.al in [9]. They presented an efficient
and effective method for hand gesture recognition. The
hand region is detected using HSV color model whrein
they applied the thresholds as 315, 94, and 37 on H, S, V
respectively through the background subtraction method.
After hand detection, segmentation was carried out to
separate out palm and fingers. Fingers and thumb were
counted to recognize the gesture. The total classification
accuracy of 1300 images tested by them has been reported
was 96.69%. However the system failed to work
satisfactorily in case of complex background.
Wei Ren Tan et.al [10] proposed a novel human
skin detection approach that combined a smoothened 2-D
histogram and Gaussian model, for automatic human skin
detection in color image(s). In their approach, an eye
detector was used to refine the skin model for a specific
person. This approach drastically reduced the
computational costs as no training was required, and it
improved the accuracy of skin detection to 90.39% despite
wide variation in ethnicity and illumination.
Log Chromaticity Color Space (LCCS) was
proposed in [11] by Bishesh Khanal et.al. which gave
illumination invariant representation of image. LCCS
resulted into an overall classification rate (CR) of about
85%. A better CR (90.45%) was obtained when LCCS was
calculated as against only luminance. In [12] , Yong Luo
et.al. removed illumination component by subtracting the
Vol. 16, No. 3, March 2018
ISSN 1947-5500

mean estimation from the original image. To make the
standardization of the overall gray values of the different
face images, ration matrix and modulus mean was
calculated and used as features. The reported recognition
rate using PCA was 92% for Yale B+ face database and
using LDA 94.28%. Hsu et. al. addressed the issue of
illumination changes, by first normalizing the image using
the geometric mean followed by a natural log of the
normalized image.[13]. The false rejection and false
acceptance ratios reported by them were as low as 0.47%
and 0% respectively.
Mohmed Alshekhali et.al. in [14] proposed the
technique for detection of hand and determination of its
center, tracking the hands trajectory and analyzing the
variations in the hand locations, and finally recognizing
the gesture. Their technique resulted in overcoming the
background complexity and gave satisfactory results for
the camera located up to 2.5 meters from the object of
interest. Experimental results indicate that this technique
could recognize 12 gestures with more than 94%
recognition accuracy.
Extensive literature review reveals that the
Luminance-Chrominance color model can be used to
detect the skin color which provides robustness against
illumination variation. Chroma (Chrominance) sampling is
the key for color based segmentation in real time
environment. YCbCr found to be promising for complex
background while HSV indicates its robustness against the
variation in the intensity of illumination while capturing
the images. In order to achieve the benefits of both YCbCr
and HSV, an approach based on the combination/fusion
of two viz. YUV (variant of YCbCr) and HSV color space
is proposed in this paper to detect the skin color. YUV
color space which was initially coded for PAL analog
video, now is also used in the CCIR 601 standard for
digital video. The detailed implementation of this fusion
and the results thereof are discussed in section III.
III. EXPERIMENTAL WORK
As discussed in section II the first clue to
segment the hand from the image is skin color. For this
purpose Luminance-Chrominance color model is used.
Pure color space (chrominance value) is used to model the
skin color; for instance UV space in YUV and SV space in
HSV color space. But under varying illumination
conditions, the skin color of the hands from different
databases, either different persons or even same person,
may vary. The sample images of hand gestures captured
under varying illumination conditions used in this paper
are shown in the Figure 2. These are available online for
research purpose and are from Sebestien Marcel database
(Figure 2.a) [15], Cambridge Hand Gesture database
(Figure 2.b) [16] and NUS Hand Posture database II
(Figure 2.c) [17].
To reduce the effects of illumination variation
effects, a normalized color space is used. Normalization is
achieved by combining YUV and HSV color spaces. For
this firstly the RGB image is converted into the YUV and
HSV color spaces using (1) to (6). This separates the
luminance and chrominance components from the image.
Separation of the chrominance approximates the
“chromaticity” of skin (or, in essence, its absorption
spectrum) rather than its apparent color value thereby
increasing the robustness against variation in illumination.
In this process, typically the luminance component is
eliminated to remove the effect of shadows, variations in
illumination etc.
Fig. 2: Images of Hand Gestures with Variation in Illumination
a) ‘Five’ from Sebestien Marcel database [15] b) ‘Flat’ from
Cambridge Hand Gesture database [16] and c) ‘B’ from NUS
Hand Posture database II [17].
YUV is an orthogonal color space in which the
color is represented with statistically independent
components. The luminance (Y) component is computed
as a weighted sum of RGB values while the chrominance
(U and V) components are computed by subtracting the
luminance component from B and R values respectively.
Mathematically this conversion is given by the following
equations:
= + 2 + (1)
= − (2)
= − (3)
HSV is invariant to dull surfaces and lighting. HSV
approximates the way humans perceive and interpret color.
The research shows that the luminance may vary due to
ambient lighting conditions and is not reliable measure to
detect the skin pixels. Saturation and Value V (brightness)
can be used in order to minimize the influence of shadow
Vol. 16, No. 3, March 2018
ISSN 1947-5500

and uneven lighting. Conversion from RGB to HSV color
space is done using following equations:
H=
⌊( ) ( )⌋
(( ) ) ( )( )
(4)
= 1 −
( , , )
(5)
= ( + + ) (6)
The same algorithm mentioned in [4] is used for human
skin detection from YUV and HSV color spaces.
Histogram is used for deciding the threshold level for
discriminating the skin and non skin pixels. The output
will be image with only skin pixels. The largest blob is
detected as hand. Arm removal algorithm is implemented
to segment the palm for further processing.
After segmenting the hand using skin color, the boundary
of the hand is detected. The object is generally described
by its boundary in a meaningful manner. Since each
boundary is composition of collection of all connected
curves, the concentration is upon the description of
connected curves. In hand gesture recognition, the
techniques which provides unique features that are used
primarily for shape representation as well as its time
complexity is less, is chosen so that the recognition of
static hand gestures can be done in real time. It is also
expected that the technique used should be invariant to
translation, rotation, and scaling.
Different methods in the literature include the use
of eccentricity, scale space and Fourier descriptors for
shape detection. 2-D Fourier transformation is extensively
used for shape representation and analysis. The details of
the literature review describing the use of Fourier
descriptors for 2-D shape detection and hand shape
detection and its implementation can be found in [18]. The
coefficients calculated by applying Fourier transform on
the input image forms the Fourier descriptors of the shape.
These descriptors generally represent the shape in a
frequency domain. The global features of the shape are
given by the low frequency descriptors and finer details of
the shape are given by the higher frequency descriptors.
The number of coefficients obtained after transformation
are generally large, some of them are sufficient to properly
define the overall features of the shape. High frequency
descriptors that are generally used to provide the finer
details of the shapes are not used for discrimination of the
shape, so they can be ignored. By doing this, the
dimensions of the Fourier descriptors used for capturing
shapes are significantly reduced and the size of feature
vector is also reduced.
As shape is connected object and is described
using a closed contour that can be represented as a
collection of the pixel coordinates in x and y direction.
The coordinates can be considered to be sampling values.
Suppose that the boundary of a particular shape has P
pixels numbered from 0 to P - 1. The pth
pixel along
boundary of the contour has position (xp, yp). The contour
can be described using two parametric equations:
x(p)= xp
y(p)= yp (7)
The Cartesian coordinates of the boundary pixel is not
considered as Cartesian coordinates instead they are
converted to the complex plane by using the following
equation:
s(p)= x(p)+ iy(p) (8)
The above equation means that the x-axis is treated as
real axis and y-axis as imaginary axis of a sequence of
complex numbers. Although the interpretation of the
sequence was recast, the nature of the boundary itself was
not changed. Of course this representation has one great
advantage: It reduces a 2-D to 1-D problem. The Discrete
Fourier Transform of this function is taken and frequency
spectra are obtained. Discrete Fourier transform of s(p) is
given by
(9)
Where u= 0, 1,2, ….P-1.
The complex coefficients a(u) are called the Fourier
descriptors of the boundary. The inverse Fourier transform
of these coefficients restores s(P) and given by the
following equation:
(10)
where p=0, 1, 2, ….P-1
To increase the robustness of the system the geometrical
features like eccentricity, aspect ratio of the area and
perimeter of the closed contour are also calculated from
the properties of the region of the hand contour. The
feature vector is formed combining the skin color based
shape features and geometrical features. The complete
algorithm of feature vector formation and classification is
represented in the following algorithm





1
0
/2
)(
1
)(
p
k
Pupj
eps
P
ua 





1
0
/2
)(
1
)(
p
u
Ppj
eua
P
PS 
Vol. 16, No. 3, March 2018
ISSN 1947-5500

Algorithm:
a) Read RGB image
b) Convert RGB to HSV and YUV
c) Apply skin detector algorithm based on the threshold
on S and U.
d) Perform the morphological operations.
e) Find the largest blob
f) Detect the palm by using arm removal algorithm.
g) Extract the boundary co-ordinates of the contour
h) Apply Fast Fourier Transform and calculate Fourier
descriptors
i) Calculate the geometrical properties of the blob
j) Combine the features (Skin color + Fourier +
Geometrical Features.) to form the feature vector.
k) Repeat the procedure for all the images in the training
and testing database.
l) Train the Backpropogation neural network to classify
the gestures.
m) Test the network and find out the accuracy.
IV.RESULTS AND DISCUSSION
As mentioned in Section III, the performance of the
system is tested using three different datasets with details
as given below.
1.Sebestien Marcel dataset consists of total 6 postures
viz. A, B, C, Point, Five and V of 10 persons in 3
different backgrounds (light, dark and complex).
2.Cambridge Hand Gesture consists of 900 image
sequences of 9 gesture classes. Each class has 100
image sequences performed by 2 subjects, captured
under 5 different illuminations and 10 arbitrary
motions. The 9 classes are defined by three
primitive hand shapes and three primitive motions.
For the experimentation we are focusing on the
hand shapes in different illumination conditions.
3.NUS Hand Posture database consists of the postures
by 40 subjects, with different ethnicities against
different complex backgrounds. The database used
in this experimentation consists of 4 hand postures
repeated five times by each of the subjects. Hand
posture images of size 160x120. 100 images are
used for training and 100 for testing.
The skin detector is first applied to extract skin regions in
the images from the three databases using fusion of HSV
and YUV color space and applying the threshold. The
results of the skin detector algorithm on the images of
three sets are presented in Fig. 4,5 and 6. Hand posture
shown in these figures are number ‘Five’ from Sebastian
Marcel dataset II, ‘B’ from NUS and ‘Flat’ from
Cambridge hand gesture dataset. The purpose of
presenting the same hand shape for all the database is to
show that the proposed system works better for complex
background and illuminations conditions. The fig. 4 shows
that the algorithm and works quiet better for Sebastian
Marcel dataset. Fig.5 represents empirical results that
show the detection of the hand region is not up to the mark
for Cambridge hand gesture database with the 5th
illumination conditions as can be seen from the Fig. 2b.
Fig. 6 interprets the result of the skin detection on the
NUS dataset. After detecting the skin, morphological
operations were performed to get the closed contour of the
hand. As explained in the algorithm in section III, the
Fourier descriptors were chosen as features and hence
were calculated from the closed contour. The descriptors
were then normalized by nullifying the 0th
Fourier
descriptor to get the invariance to the translation. Scale
invariance was obtained by dividing all Fourier descriptors
by the magnitude of the 1st
Fourier descriptor. Rotation
invariance is achieved by only considering the magnitude
of the Fourier coefficients.
The feature vector was formed by considering 20
coefficients of Fourier descriptors (which are invariant to
scale, rotation and translation) and two geometrical
features viz. area to perimeter aspect ratio and eccentricity,
thus making a total of 22 features. Geometrical features
were calculated from the closed contour of the segmented
hand. The feature vectors thus formed were then used to
train and test the multilayer feed forward neural network
to classify the hand gesture. For learning the network,
Back propagation algorithm with Levenberg-Marquardt
algorithm has been used to train the network. The
activation function used is “Sigmoid”. Fig. 3 shows the
architecture of the NN used in this experiment.
Fig. 3. Neural Network Architecture
The same experiment is tested for NUS Hand Posture
dataset I which consists of 10 classes of postures with 24
samples of each. As there is uniform background the
classification accuracy observed is 100%.
Vol. 16, No. 3, March 2018
ISSN 1947-5500

Fig. 4. Results of skin detector -Sebastian Marcel dataset II.
Fig. 5. Results of skin detector for posture ‘flat’ from the
Cambridge dataset
Fig . 6. Result of skin detector for posture ‘B’ from NUS Hand
Posture Dataset II.
The results of the proposed work are presented in the
following tables. For the experimentation six postures ‘A’,
‘B’, ‘C’, ‘point’, ‘V’, ‘Five’ from Sebastien Marcel
database has been used. Table 1 describes the individual
accuracy for each of these postures. The average accuracy
achieved is 96%.
TABLE 1: CLASSIFICATION ACCURACY FOR SEBESTIEN MARCEL
DATABASE
Static
gesture
No. of
gesture
samples
Correct Incorrect Classification
Accuracy %
A 100 96 04 96
B 100 94 6 94
C 100 95 5 95
Point 100 98 2 98
V 100 96 4 96
Five 100 97 3 97
Avearage Classification Accuracy 96
The proposed work is compared with the existing state of
art techniques on the same benchmark dataset in Table 2
which expose that the experiment conducted in this paper
is comparable with those of existing techniques.
TABLE 2: COMPARISON WITH EXISTING STATE OF ART TECHNIQUES FOR
SEBASTIAN MARCEL DATABASE
Paper No. Features Classifier Accuracy
(%)
[19] Modified Cesnsus
Transform
AdaBoost 81.25
[20] Haar like features AdaBoost 90.0
[21] Haar wavelets Penalty score 94.89
[22] Scale space features AdaBoost 93.8
[23] Bag of features Support Vector
Machine
96.23
[24] Normalized
Moment of Inertia
(NMI) and Hu
invariant moments
Support Vector
Machine
96.9
Proposed
method
Skin color and
Fourier Descriptor
Artificial
Neural Network
96
Four hand gestures ‘A’, ‘B’, ‘C’, ‘D’ are used for the
experiments from the NUS Hand Posture dataset. 100
samples for each posture are used for training and 100 are
used for testing. The results of the experiment are
presented in Table 3. The average accuracy of 95.25% is
achieved.
Vol. 16, No. 3, March 2018
ISSN 1947-5500

TABLE 3. CLASSIFICATION ACCURACY FOR NUS HAND POSTURE
DATABASE II
Static
gesture
No. of
gesture
samples
Correct Incorrect Classification
Accuracy %
A 100 96 4 94
B 100 94 6 94
C 100 97 3 97
D 100 96 4 96
Avearage Classification Accuracy 95.25
The results obtained through this experimentation are
compared with the state of art techniques. The comparison
reveals that the proposed method is better than the existing
methods. The details of this are given in the table 4.
TABLE 4. COMPARISON WITH EXISTING STATE OF ART TECHNIQUES FOR
NUS HAND POSTURE DATABASE II
Paper
No.
Features Classifier Accuracy
[25] Shape based and
texture based
features
GentleBoost 75.71
[26] Viola jones Real Time
Deformable
Detetctor
90.66
[27] NUS
standard model
features (SMFs)
Fuzzy Rule
Classifier
Support Vector
Machine
93.33
92.50
[28] Shape texture
color
Support Vector
Machine
94.36
Propos
ed
method
Skin color and
Fourier Descriptor
Artificial Neural
Network
95.25
Three primitive hand shapes ‘flat’, ‘Spread’ and
‘V’ from Cambridge Hand gesture database are used for
testing the proposed algorithm. The results of the proposed
work are presented in the Table 5. The experiment is
carried out for each set of database and reported in the
table. The average accuracy is 93.67% .
The results obtained through this
experimentation are compared with the state of art
techniques and reported in Table 6.
V.CONCLUSION AND FUTURE SCOPE
The paper proposed a system for hand
segmentation and classification. The main component of
the system is to track the hand based on skin color under
different illumination conditions and with complex
background. Fusion of HSV and YUV color space to
detect the skin color gave the invariance to the
illumination even in the complex background. The closed
contour of the segmented hand is used to detect the shape
of the hand gesture. Fourier descriptors are calculated as
shape descriptors. To improve the robustness for the shape
detection, the geometrical features are added in the feature
vector. The feature vector thus achieved by combining the
shape features and geometrical features are given to the
artificial neural network for classification. The average
classification accuracy of 95.25% is achieved for all the
three databases.
The hand postures in the databases have the
different viewing angle. So the classification accuracy can
be further increased by extracting the view invariant
features from the images. This lays a direction for further
reseach in this area.
TABLE 5. CLASSIFICATION ACCURACY FOR CAMBRIDGE HAND GESTURE
DATABASE
Static
gesture
No. of
gesture
samples
Set 1 Set2 Set 3 Set 4 Set 5
Flat 100 93 96 96 92 94
Spread 100 94 95 93 93 93
V 100 95 96 95 92 94
Avearage
Classification
Accuracy
94 95.67 94.67 92.33 93.67
TABLE 6. COMPARISON WITH EXISTING STATE OF ART TECHNIQUES FOR
CAMBRIDGE HAND GESTURE DATABASE
Paper No. Features Classifier Accuracy
(%)
[29] PCA on Motion
gradient orientation
Sparse
Bayesian
Classifier
80
[30] Canonical
Correlation Analysis
(CCA) + SIFT
Support Vector
Machine
85
[31] Concatenated HOG Kernel
Diceriminant
analysis with
RBF kernel
91.1
[32] Fourier Descriptors
(Static postures 4
shapes)
Support Vector
Machine
92.5
Proposed
method
Skin color and
Fourier Descriptor
ANN 94.50
REFERENCES
[1]. Vladimir I. et.al. ‘Visual Interpretation of Hand Gestures for
Human-Computer Interaction: A Review’, IEEE transactions on
Pattern Analysis and Machine Intelligence, vol. 19, no. 7, July
1997.
[2]. Sushmita Mitra, Tinku Acharya, ‘Gesture Recognition: A Survey’
IEEE Transactions on Systems, Man, and Cybernetics—Part C:
Applications and Reviews, Vol. 37, No. 3, MAY 2007.
Vol. 16, No. 3, March 2018
ISSN 1947-5500

[3]. Terrillon, J., Shirazi, M. N., Fukamachi, H., Akamatsu, S,
‘Comparative performance of different skin chrominance models
and chrominance spaces for the automatic detection of human faces
in color images’, IEEE International Conference on Face and
Gesture Recognition, pp. 54-61 (2000)
[4]. Khamar Basha Shaika et.al ‘Comparative Study of Skin Color
Detection and Segmentation in HSV and YCbCr Color Space’
International Elsevier Conference on Recent Trends in Computing
2015
[5]. Zhang Qiu-yu, Lu Jun-chi, et.al, “Hand Gesture Segmentation
Method Based on YCbCr Color Space and K
International Journal of Signal Processing, Image Processing
Pattern RecognitionVol. 8, No. 5 (2015), pp. 105
[6]. Bahare Jalilian, Abdolah Chalechale , ‘Face and Hand Shape
Segmentation Using Statistical Skin Detection for Sign Language
Recognition’, Computer Science and Information Technology
.DOI: 10.13189/csit.2013.010305
[7]. Hsiang-Yueh. Lai 1, Han-Jheng. Lai, “Real
Gesture Recognition”, IEEE DOI 10.1109/IS3C.2014.177
Published in: Computer, Consumer and Control (
International Symposium
[8]. Nasser H. Dardas and Emil M. Petriu,” Hand Gesture Detection and
Recognition Using Principal Component”, IEEE International
Conferenc Computational Intelligence for Measurement Systems
and Applications (CIMSA), 2011
[9]. Zhi-hua Chen,1 Jung-Tae Kim,1 Jianning Lian
Gesture Recognition Using Finger Segmentation” The Scientific
World Journal Volume 2014 (2014), Article ID
http://dx.doi.org/10.1155/2014/267872
[10]. Wei Ren Tan, et.al., “A Fusion Approach for Efficient Human Skin
Detection”, IEEE transaction on Industrial informatics Vol8(1)
PP138-147
[11]. Bishesh Khanal, et.al “Efficient Skin Detection Under Severe
Illumination Changes and Shadows”. ICIRA 2011
International Conference on Intelligent Rob otics and Applications,
Dec 2011, Aachen, Germany. pp.1-8, 2011.
[12]. Luo et.al. “Illumination Normalization Method Based on Mean
Estimation for A Robust Face Recognition” Research article.
[13]. Hsu, R., Ab del-Mottaleb, M., Jain, A. K.: Face detecting in color
images. IEEE Trans. on Pattern Analysis and Machine Intelligence
24, 696-706 (2002)
[14]. Mohamed Alsheakhali, Ahmed Skaik “Hand Gesture Recognition
System”, International Conference on Information and
Communication Systems (ICICS 2011)
[15]. S. Marcel, “Hand posture recognition in a body
space,” in Proc. Conf. Human Factors Comput. Syst. (CHI), 1999,
pp. 302Processing, 2005, vol. 2, pp. 954 – 957
[16]. T-K. Kim et.al, “ Tensor Canonical Correlation Analysis for Ac
Classification “,In Proc. of IEEE Conference on Computer Vision
and Pattern Recognition (CVPR), Minneapolis, MN, 2007.
[17]. Pramod Kumar Pisharady, Prahlad Vadakkepat, Ai Poh Loh,
"Attention Based Detection and Recognition of Hand
Postures Against Complex Backgrounds", International Journal of
Computer Vision, vol.101, no.3, pp.403-419, February, 2013
[18]. Mrs.A. R.Patil, Dr. S. Subbaraman, “Static Hand Gesture Detecti
and Classification using Contour based Fourier Descriptor”, 9th
World Congress International conference on Science and
Engineering Technology at GOA. Jan. 2016
://iferp.in/digital_object_identifier/WCASETG9718
[19]. Agn`es Just et.al.,’ Hand Posture Classification and Recognition
using the Modified Census Transform’, 7th IEEE International
Conference on Automatic Face and Gesture Recognition (FGR’06)
0-7695-2503-2/© 2006 IEEE
[20]. Q. Chen, N. Georganas, and E. Petriu, “Real
hand gesture recognition using Haar-like features,” in Proc. IEEE
IMTC, 2007,pp. 1–6.
[21]. Y. Fang, K. Wang, J. Cheng, and H. Lu, “A real
recognition method,” in Proc. IEEE Int. Conf. ultimedia Expo,
2007, pp. 995–998.
Terrillon, J., Shirazi, M. N., Fukamachi, H., Akamatsu, S,
‘Comparative performance of different skin chrominance models
for the automatic detection of human faces
in color images’, IEEE International Conference on Face and
Khamar Basha Shaika et.al ‘Comparative Study of Skin Color
Detection and Segmentation in HSV and YCbCr Color Space’ , 3rd
International Elsevier Conference on Recent Trends in Computing
chi, et.al, “Hand Gesture Segmentation
Method Based on YCbCr Color Space and K-Means Clustering”,
International Journal of Signal Processing, Image Processing and
Pattern RecognitionVol. 8, No. 5 (2015), pp. 105-1 16
Bahare Jalilian, Abdolah Chalechale , ‘Face and Hand Shape
Skin Detection for Sign Language
Recognition’, Computer Science and Information Technology
Jheng. Lai, “Real-Time Dynamic Hand
DOI 10.1109/IS3C.2014.177
Computer, Consumer and Control (IS3C), 2014
Nasser H. Dardas and Emil M. Petriu,” Hand Gesture Detection and
Recognition Using Principal Component”, IEEE International
putational Intelligence for Measurement Systems
Jianning Lian “Real-Time Hand
Gesture Recognition Using Finger Segmentation” The Scientific
(2014), Article ID 267872,
Wei Ren Tan, et.al., “A Fusion Approach for Efficient Human Skin
n”, IEEE transaction on Industrial informatics Vol8(1)
Bishesh Khanal, et.al “Efficient Skin Detection Under Severe
Illumination Changes and Shadows”. ICIRA 2011 - 4th
International Conference on Intelligent Rob otics and Applications,
Luo et.al. “Illumination Normalization Method Based on Mean
Estimation for A Robust Face Recognition” Research article.
Mottaleb, M., Jain, A. K.: Face detecting in color
s and Machine Intelligence
Mohamed Alsheakhali, Ahmed Skaik “Hand Gesture Recognition
International Conference on Information and
S. Marcel, “Hand posture recognition in a body-face centered
pace,” in Proc. Conf. Human Factors Comput. Syst. (CHI), 1999,
957
Tensor Canonical Correlation Analysis for Action
“,In Proc. of IEEE Conference on Computer Vision
and Pattern Recognition (CVPR), Minneapolis, MN, 2007.
Pisharady, Prahlad Vadakkepat, Ai Poh Loh,
Attention Based Detection and Recognition of Hand
", International Journal of
419, February, 2013
Mrs.A. R.Patil, Dr. S. Subbaraman, “Static Hand Gesture Detection
and Classification using Contour based Fourier Descriptor”, 9th
World Congress International conference on Science and
Engineering Technology at GOA. Jan. 2016 http
WCASETG9718
Hand Posture Classification and Recognition
, 7th IEEE International
Conference on Automatic Face and Gesture Recognition (FGR’06)
and E. Petriu, “Real-time vision-based
like features,” in Proc. IEEE
Y. Fang, K. Wang, J. Cheng, and H. Lu, “A real-time hand gesture
recognition method,” in Proc. IEEE Int. Conf. ultimedia Expo,
[22]. W. Chung, X. Wu, and Y. Xu, “A real time hand gesture
recognition based on Haar wavelet representation,” in Proc. IEEE
Int. Conf. Robot. Biomimetics, 2009, pp. 336
[23]. Nasser H. Dardas and Nicolas D. Georganas, “Real
Gesture Detection and Recognition Using Bag
Support Vector Machine Techniques”. IEEE Transactions on
Instrumentation and Measurement, vol.60, no.11, November 2011
[24]. Y. Ren and C. Gu, “Real-time hand gesture recognition based on
vision,” in Proc. Edutainment, 2010, pp. 468
[25]. Serre, T., et.al.(2007), ‘Robust object recognition with cortex
mechanisms’, IEEE Transactions on Pattern Analysis and Machine
Intelligence, 29(3), 411–426
[26]. Uriel Haile Hernandez-Belmonte and Victor Ayala
Time Hand Posture Recognition for Human
Tasks’, Sensors 2016, 16, 36;
www.mdpi.com/journal/sensors
[27]. P. Pramod Kumar et.al,’ Hand Posture And Face Recognition Using
A Fuzzy- Rough Approach
Robotics Vol . 7, N o. 3 (2010) 331
Publishing Company DOI: 10.1142/S0219843610002180
[28]. Pramod Kumar Pisharady et.al,
Recognition of Hand Postures
Against Complex Backgrounds
10.1007/s11263-012-0560-5 published by springerlink
[29]. Shu-Fai Wong and Roberto Cipolla, ‘ Real
Hand Motions using a Sparse Bayesian Classifier on Motion
Gradient Orientation Images’
[30]. T.-K. Kim and R. Cipolla. Canonical correlation analysis of video
volume tensors for action categorization and detection. IEEE Trans.
Pattern Analysis and Machine Intelligence, 31(8):1415
[31]. Masoud Faraki Mehrtash T. Harandi Fatih Porikli ,
Classification by Symmetric Positive Semi
[32]. Heba M.Gamal et.al, ‘ Hand Gesture Recognition using Fourier
Descriptors’, 978-1-4799-008 274 0
Authors’ Profiles
Mrs.Anjali R Patil have completed BE
(2002)and ME(2009) from Shivaji
University, Kolhapur. Currently she is
Pursuing PhD in Electronics Engineering
from Shivaji University Kolhapur. Her
research interest includes image
processing, Pattern rec
computing etc.
Dr. Mrs. Shaila Subbaraman, Ph. D. from
I.I.T., Bombay (1999) and M. Tech. from
I.I.Sc., Bangalore (1975) has a vast
experience in industry in the capacity of R
& D engineer and Quality Assurance
Manager in the field of man
semiconductor devices and ICs. She also has more than 27 years
of teaching experience at both UG and PG level for the courses
in Electronics Engineering. Her specialization is in Micro
electronics and VLSI Design. She has more than fifty
publications to her credit. She retired as Dean Academics of
autonomous Walchand College of Engineering, Sangli and
currently she is working as Professor (PG) in the same college.
Additionally she works as a NBA expert for evaluating
engineering programs in India
Accord. Recently she has been felicitated by “Pillars of
Hindustani Society” award instituted by Chamber of Commerce,
Mumbai for contribution to Higher Education in Western
Maharashtra.
W. Chung, X. Wu, and Y. Xu, “A real time hand gesture
based on Haar wavelet representation,” in Proc. IEEE
Biomimetics, 2009, pp. 336–341
Nasser H. Dardas and Nicolas D. Georganas, “Real-Time Hand
tion and Recognition Using Bag-of-Features and
Support Vector Machine Techniques”. IEEE Transactions on
Instrumentation and Measurement, vol.60, no.11, November 2011
time hand gesture recognition based on
nt, 2010, pp. 468–475
Robust object recognition with cortex-like
, IEEE Transactions on Pattern Analysis and Machine
426
Belmonte and Victor Ayala-Ramirez ,’Real-
Posture Recognition for Human-Robot Interaction
, Sensors 2016, 16, 36; doi:10.3390/s16010036
www.mdpi.com/journal/sensors
Hand Posture And Face Recognition Using
Rough Approach’, International Journal of Humanoid
Vol . 7, N o. 3 (2010) 331–356 World Scientific
Publishing Company DOI: 10.1142/S0219843610002180
Pramod Kumar Pisharady et.al, ‘Attention Based Detection and
Recognition of Hand Postures
t Complex Backgrounds’,Journal Computer Vision DOI
5 published by springerlink
Fai Wong and Roberto Cipolla, ‘ Real-time Interpretation of
Hand Motions using a Sparse Bayesian Classifier on Motion
Gradient Orientation Images’, BMVC 2005 doi:10.5244/C.19.41
K. Kim and R. Cipolla. Canonical correlation analysis of video
volume tensors for action categorization and detection. IEEE Trans.
Pattern Analysis and Machine Intelligence, 31(8):1415–1428, 2009.
Masoud Faraki Mehrtash T. Harandi Fatih Porikli , ‘Image Set
Classification by Symmetric Positive Semi-Definite Matrices’
Hand Gesture Recognition using Fourier
008 274 0-0/13/$31.00 ©2013 IEEE
Mrs.Anjali R Patil have completed BE
(2002)and ME(2009) from Shivaji
University, Kolhapur. Currently she is
Pursuing PhD in Electronics Engineering
from Shivaji University Kolhapur. Her
research interest includes image
processing, Pattern recognition, Soft
Dr. Mrs. Shaila Subbaraman, Ph. D. from
I.I.T., Bombay (1999) and M. Tech. from
I.I.Sc., Bangalore (1975) has a vast
experience in industry in the capacity of R
& D engineer and Quality Assurance
Manager in the field of manufacturing
semiconductor devices and ICs. She also has more than 27 years
of teaching experience at both UG and PG level for the courses
in Electronics Engineering. Her specialization is in Micro-
electronics and VLSI Design. She has more than fifty
tions to her credit. She retired as Dean Academics of
autonomous Walchand College of Engineering, Sangli and
currently she is working as Professor (PG) in the same college.
Additionally she works as a NBA expert for evaluating
engineering programs in India in accordance with Washington
Accord. Recently she has been felicitated by “Pillars of
Hindustani Society” award instituted by Chamber of Commerce,
Mumbai for contribution to Higher Education in Western
Vol. 16, No. 3, March 2018
ISSN 1947-5500

Illumination Invariant Hand Gesture Classification against Complex Background using Combinational Features

More Related Content

What's hot (19)

Similar to Illumination Invariant Hand Gesture Classification against Complex Background using Combinational Features (20)

Recently uploaded (20)

Illumination Invariant Hand Gesture Classification against Complex Background using Combinational Features