INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
SCIENCE, ENGINEERING AND TECHNOLOGY
IJSET is a scholarly peer-reviewed journal that contributes for the contemporary trends in the
field of Science, Engineering and Technology published quarterly by Vankatram Learning Centre
(Central Library), Sri Krishna College of Engineering and Technology. Responsibility for the
contents rests upon the authors and not upon the IJSET. For copying or reprint permission, write
copyright department, IJSET, Vankatram Learning Centre, Sri Krishna College of Engineering and
Technology, Kuniamuthur Post, Coimbatore – 641 008, Tamil Nadu, India.
Patron
Smt.S.Malarvizhi
Advisor
Dr.S.Sundararaman
Editor in Chief
Dr.S.Annadurai
Editor
Dr. A.F.Rahman
Managing Trustee
Chief Executive Officer
Principal
Librarian
Editorial Board
1. Dr.K.Duraisamy
Professor, KSR College of Tech., Tiruchengode
2. Dr.L.Ganesan
Professor, Alagappa Chettiar College of
Engg. & Tech., Kariakudi, Tamilnadu
3. Dr.S.Thamarai Selvi,
MIT, Dean CT, Anna University, Chennai
4. Dr.R. Shanmugalakshmi
Associate Professor, GCT, Coimbatore
5. Dr.T.Purusothaman
Associate Professor, GCT, Coimbatore
6. Dr.V. Gopalakrishnan
Associate Professor, GCT, Coimbatore
7. Dr.Narasiman
Professor, Alagappa Chettiar College of
Engg. & Tech., Kariakudi, Tamilnadu
8. Dr.Esther Annlin Kala James
Principal, GCE, Srirangam, Trichy
9. Dr.P.Geetha
1. Prof. Palesi Maurizio
Associate Professor
University of Enna “Kore”, Italy
2. Prof. Leo Mathew,
Professor, Texas University, Austin
3. Dr. Selwyn Piramuthu
Professor, University of Florida, USA
4. Dr.Henry Selvaraj
Professor, University of Nevada, USA
5. Dr. Sameer Hanna Khader
Associate Professor,
Palestine Polytechnic University, Palestine
6. Prof. Vijay K Arora
Professor, Wilkes University, USA
7. Dr.Kevin Demmitt,
Provost and Vice President,
Clayton State University, Georgia
8. Prof. Scott E.Mclntyre,
Professor, University of TOLEDO, USA
Associate Professor, Anna University, Chennai
9. Dr. Hairul Azhar Abdul Rashid
Professor, Sri Ramakrishna Engg. College, CBE
Director Research Management Centre,
Multimedia University, Malaysia
10. Dr.Grace Selvarani
11. Dr.Anna Saro
Director and Head, Dept. of MCA,
SNR Sons College, CBE
10.Dr. Teerapat Sanguankotchakorn
Associate Professor
Asian Institute of Technology, Thailand
12. Dr.Arulanandan
INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
Professor, PSG Tech, Coimbatore
SCIENCE, ENGINEERING AND TECHNOLOGY
INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
SCIENCE, ENGINEERING AND TECHNOLOGY
INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
SCIENCE, ENGINEERING AND TECHNOLOGY
International Journal on Advanced Research in Science, Engineering and Technology is
published as a quarterly publication.
The journal is a scholarly peer-reviewed journal that contributes for the contemporary trends
in the field of Science and Technology.
IJSET or editorial board accepts no responsibility for the statements and opinions advanced by
contributors. The Editorial Committee’s decision regarding suitability of contributions for
publication will be final.
All correspondence and manuscripts of contributions should be sent to
Dr.S.Annadurai
Principal & Editor-in-Chief (IJSET)
International Journal on Advanced Research
in Science, Engineering and Technology
Sri Krishna College of Engineering and Technology
Kuniamuthur Post
Coimbatore : 641 008
Tamil Nadu, India.
Marking a copy to
Dr.A.Fazlur Rahman
Librarian & Editor (IJSET)
International Journal on Advanced Research
in Science, Engineering and Technology
Vankatram Learning Centre
Sri Krishna College of Engineering and Technology
Kuniamuthur Post
Coimbatore : 641 008
Tamil Nadu, India.
E-Mail : librarian@skcet.ac.in
drfrlib@gmail.com
© 2016, IJSET, Vankatram Learning Centre, Sri Krishna College of Engineering and
Technology, Coimbatore.
INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
SCIENCE, ENGINEERING AND TECHNOLOGY
MAY – AUGUST 2016
VOLUME 1
NUMBER 1
IJARSET
(ISSN
)
CONTENTS
Sl.No.
Title
Page
No.
1
A Novel Approach for Identifying Common Signs of CT Pathological Lung
Images………………………………...………… T. Revathi, and P. Geetha
01
2
Ensemble Approach on Maize Leaf Nutrition Deficiency Identification Using
Enhanced Multivariate Partial Least Square…………………………………..
……………………………………….. S. Sridevy, and Anna Saro Vijendran
09
3
Low Power Multiplier Using Algorithmic Noise Tolerant Architecture
.…………………………………………………...... S. Fathima, and S. Uma
16
4
A Survey on Compression in Encrypted Domain …………………………...
……………………………………….. T. Suguna, and R. ShanmugaLakshmi
20
5
A Survey on Assistive Technology using Natural User Interface(NUI)
computing to support and augment therapeutic rehabilitation ………………..
.. ...…… S. J. Syed Ali Fathima, S. Shanker, and A. Ahamed Thajudeen
27
6
Experimental Investigation on Weld Strength of Copper Wire – Al8011 Sheet
Using Ultrasonic Metal Welding .....................................................…………
…………………. R. Karthik, P. Ashoka Varthanan, and V. S. Gowthaman
32
7
Effect of Vibratory treatment on Metallurgical and Mechanical properties of
Aluminum alloys weldments……… K. Balasubramanian, and V. Balusamy
40
A Novel Approach for Identifying Common Signs of
CT Pathological Lung Images
T. Revathi
Dr.P.Geetha
PG Student
Dept. of Computer Science & Engg
College of Engineering, Guindy
Anna University, Chennai.
revathiarasu.t@gmail.com
Associate Professor
Dept. of Computer Science & Engg
College of Engineering, Guindy
Anna University, Chennai.
geetha@cs.annauniv.edu
Abstract-Accurate identification of pathological lungs
& its sign is a challenging task in medical imaging
applications. The existing methods fail to provide a better
solution for images with dense pathologies and they
falsely exclude some regions from the lung parenchyma.
To tackle this problem, a novel pathological identification
system is required. It should work in a way that it
encounters almost all common pathologies & its sign
present in the image set. This proposed approach is fully
based on Volume based pathologies identification. It has
two stages. Stage one is for Volume detection & Volume
difference
test,
where
Fuzzy-Connectedness(FC)
segmentation and morphological - Rib cage Volume
detection method is used. FC is used for segmenting lungs
with diverse range of lung abnormalities. Refinement
steps are conducted in stage two by using Random Forest
(RF) algorithm which further refines the images such as
pathological existence and its sign by using textural
information of images. It classifies the lung images into
two main classes namely pathological class and nonpathological class. Pathological class has nine sub
categories such as GGO(Ground Grass Opacity),
Lobulation, Air Bronchogram, Cavity, Plueral
Indentation(PI), Spiculation, Calcification, Obstructive
Pneumonia and Bronchial Mucus Plugs. So these
subcategories of signs are also identified for effective
diagnosis of lung diseases.
Some wide variety of imaging techniques are available today
such as magnetic resonance imaging (MRI), computed
tomography (CT), etc. But interpreting a CT image is very
challenging because the image is complicated by
superimposed anatomical structures. Also some current lung
pathological identification algorithms work well for certain
lung pathologies present in moderate amounts, they fail to
perform when dense pathologies exist. Accurate
segmentation of the pathological lung is challenging since
lung pathologies hold different appearances from the normal
lung tissues. And CT scans of patients with severe lung
diseases often include diverse imaging patterns. Thus, it is
difficult to adapt the current methods due to their limited
ability to be applied on finding diverse imaging patterns with
dense pathologies. However misdiagnosis the image will
affect patient care and treatment. So, further studies are
necessary in the development of medical imaging software to
produce more details of the image and diseases. Therefore the
aim of this work is to target this challenge and provide a
generic solution for lung pathology & its sign identification
from CT scans.
There were so many diseases identification
techniques had been developed, but each approach
concentrates on particular diseases detection like respiratory
problem, lung diseases, pulmonary diseases, nodule diseases
& pleural effusion diseases etc. Mostly earlier studies were
concentrated on lung cancer detection. There is no such
generic approach which finds all the lung abnormalities at a
single stretch. Therefore accurate segmentation of the
pathological lung is challenging. So, there is a need for such
approach to recognize all the lung abnormalities in a single
system.
Index Terms—Lung Segmentation, Fuzzy Connected
-ness, Random Forest.
I.
INTRODUCTION
Medical imaging plays an important role in medical
world nowadays. Pulmonary Lung diseases and disorders are
one of the major causes of deaths and hospitalization around
the world. The American Lung Association estimates that
about 90% of deaths occur per year in the United States from
lung diseases[2].
Computed tomography is the current modality used in
clinics for detecting and diagnosing the lung diseases[2].
Since manual inspection is tiresome and time consuming.
II.
LITERATURE SURVEY
The study of various lung segmentation techniques and
classification techniques are described in below sections.
A. Lung Segmentation Techniques
The aim of medical image segmentation is to extract
quantitative
information
(e.g.,volumetric
data,
morphological data, textural information) with regard to an
1
organ of interest or a lesion within the organ. Some major
lung segmentation methods are Threshold-based, Regionbased Segmentation methods.
Threshold based segmentation method segments the
images or distinguishes the image regions with respect to
contrast / intensities present in the images. Threshold - based
methods [2], [3] are often used for their efficiency. However,
such methods have limited applicability as they fail to
consider the intensity variations due to pathologies presence
or even under normal conditions. In [4] used multi-level
thresholding for identifying the regions of interest (ROI) of
lung nodule. 5% –17% of the lung nodules are undetected.
In[5], Adaptive threshold method is used for segmenting the
lung region. It fails to deal with attenuation variations.
Images with non uniformity and noise presence leads to
misclassification.
Region Based Segmentation method overcomes the
drawback of threshold based segmentation. It segments
images based on the regions with similar characteristics. It
has 2 Methods: Region growing method, Region Splitting &
Merging. It is used in[6], on which neighboring pixels are
scanned and added to a region session. In [7][8],
Combination of spatial Fuzzy C-Means and morphological
techniques for segmenting the lung region from CT lung
images. They have divided their work into three main stages
namely
Background
removal,
Preprocessing,
and
Morphological based operations. Among them, a new
histogram background removal operator had been used to
remove the pixels around the lungs. Moreover, preprocessing
is performed by smoothing and removing the noise presents
in the segmented image. Finally, morphological operations
are used to separate the edges and to fill the small holes
present in the segmented lungs. In[9], ROI based region
segmentation, the initial step in computer aided diagnosis of
lung CT image is generally to segment the Region of Interest
(ROI) present in the image and then analyzes each area
separately in-order to find the presence of pathologies present
in ROI. Region growing has been combined with CC in this
work since it reduces the number of steps in segmentation for
the process of identifying a tissue in the CT lung image.
In[8], The lung lobes and nodules in CT image are segmented
using two stage approaches such as modified adaptive
fissure sweep and adaptive thresholding. Initially preprocessing is used to remove the noise present in CT image
using filter technique, then the fissure regions are located
using adaptive fissure sweep technique, then histogram
equalization and region growing is applied to refine the
oblique fissure.
Therefore the above techniques were only concentrated
on identifying particular diseases or disorders. It doesn’t
identify all pathological signs or diseases in a single system.
Even it failed to deal with attenuation variations & noise
presence in the image. Region growing segmentations covers
all the neighbor pixels which has same characteristics, this
mean that it covers almost all pixels with same intensity.
Whereas the new Fuzzy Connectedness Approach finds the
neighbor pixel which has similar spatial & textural property
by means of calculating affinity value. FC approach works
well even the images consists of noise.
B. Classification Based On Machine Learning
It focuses on extracting suitable features such as shape or
texture for a predefined classifier such as support vector
machines, random forests, neural networks etc. Support
Vector Machine(SVM)
performs classification by
constructing N dimensional hyper plane that optimally
separates the data into two categories. It is a linear classifier
and this is not suitable when giving large number of datasets
or continuous data. Decision trees are predictive models that
use a set of binary rules to calculate a target value. In[12],
addresses the challenging problem of segmenting the lungs in
CT scans. They proposed the context selective decision forest
(CSDF) as a new discriminative classifier which resulting in
higher prediction accuracy and greater use for the clinic.
Quantitative comparisons were made with state of art
algorithms and demonstrated comparable accuracy with
superior computational efficiency. Benefits of using decision
forest is it can able to handle continuous data and large
number of features. It provides more correctness and
accuracy comparatively. It is easy to implement and less time
consuming.
This Survey clearly shows the segmentation techniques
with its benefit and drawback. From this, we infer that, first
order segmentation such as Threshold, Region, Edge are not
suitable due to its flaw. So Fuzzy Connectedness(FC) will be
used to segment the lung and also to get volume information.
Whereas in Classification, Decision Forest is the most
suitable one to handle large number of features as well the
continuous data. So Random Forest would be the best for
Classification work since it combines many decision trees
and provides higher accuracy and performance.
III.
ARCHITECTURE
Fig.1 illustrates the entire system of pathological lung
identification approach. Input of the system is Lobe and Lung
Analysis (LOLA) images. Fuzzy Connectedness based
segmentation is performed on the input image to get
segmented image with its volume. Then for Pathological
testing, volume difference test will be performed. For that it
needs two Volumes. Volume one was got from FC
segmentation (ie., Lung Volume), In order to find second
volume of the image, Morphological Rib Cage analysis
technique is used. Then Rib Cage is determined using
Morphological
operations
by
removing
scapula.
2
Fig.1. Proposed Architecture
Because Scapula presence will results in getting increased
volume of the image. Then Convex hull technique is used to
fit polygon over the Rib cage identified. This process has to
be carried out for each and every image to find Volume of
Convex hull fitted Rib image. Finally Volume difference test
will be performed over 2 volumes, one got from FC
Segmentation and other from Rib Cage Convex Hull. From
this we infer that the image has presence of pathology or not.
The image is said to be Pathological image if Volume
difference occurs or else the images is said to be normal/
non-pathological image. Subcategories of pathological class
is given in fig.9. In-order to confirm and refine the results
obtained from Pathology Test, it moves to multi stage
refinement level where Random Forest Classifier is used. It
takes Training dataset which is collected from wide range of
pathological images & signs of pathological images such as
GGO(Ground Grass Opacity), Calcification, Cavity,
Spiculation, Lobulation, Air Bronchogram, Bronchial Mucus
Plugs, Plueral Identation(PI) and Obstructive Pneumonia[11].
The test dataset contains the feature extracted from LOLA
datasets. It models the classifier based on the Training set and
performs classification by means of passing the test sets over
it. Final output of the system gives 2 main classes of images,
class 0 which has non-pathological images and class 1 which
has pathological images with its sign.
IV.
MODULES DESCRIPTION
Description of modules such as Fuzzy Connectedness,
Convex Hull, Random Forest are described below.
3
b) Strength is determined by all possible
connecting paths between those pixels.
c) Each pair of nearby pixel has affinity value.
d) It traverses through the pixel with minimum
affinity value.
e) When some abnormality is seen, FC stops
region growing and outputs the segmented area. (Initial Seed/
pixel Selection can be done either manually or
automatically). Output of segmented lung is given in fig.6.
Step 1) FC Algorithm
Input: Image and Seed point.
Output: Segmented Image.
1. Read Image.
2. Convert to gray scale.
3. Select Seed Point /Initial pixels (say c,d).
4. Find Adjacency coordinates of pixels.
A. Segmentation Based On Fuzzy Connectedness
It is the representation of connectedness between the
pixels comprising an object in an image. It is developed
mainly for medical image segmentation. The seed / initial
pixel is determined by means of computation. Sometimes
multiple seeds also to be considered.
Using an initial seed pixel within an object in an image,
it computes affinity value for the adjacent pixels & traverse
through it. Here intensive computation may be required
because connectedness is defined based on an optimal path to
the seed pixel.
Dynamic programming is used to determine the optimal
paths. Sometimes operator-selected region of interest has also
been applied to reduce computation time. Fuzzy
connectedness is closely related to other graph-search
algorithms. Final object selection is performed using a
threshold on the resulting fuzzy connectedness map.
It has three components namely Homogeneity, Fuzzy &
Object Spel Adjacency.
Homogeneity based component indicates the
degree of local hanging togetherness of spels
due to their similarity of intensities.
Fuzzy Spel Adjacency component indicates the
degree of spatial adjacency of spels.
Object Spel Adjacency component indicates the
degree of local hanging togetherness of spels
with some given object feature.
(1)
5. Computes Affinity
Let fuzzy affinity Psi (c, d) quantify the
hanging-togetherness of two spels c and d.
(2)
6. Do Absolute Fuzzy Connectedness to get FC map
For any path π, the strength of the path is
minimum affinity along the path.
(3)
Find Maximum Path:
(4)
5. Return FC connectivity map
6. Adjust threshold if necessary
7. Get best Segmented Lung.
Step 2) Maximum Path Coverage Algorithm
1. Initialize array with fc=0:1.
2. Push all pixels μ (c, d) > 0 to Q.
3. While Q is not empty do
a. Find strongest fc from Q.
b. Find its adjacency & remove it.
c. Find fmax= path with minimum affinity.
If fmax > fc then
i. Set max f c =fmax.
ii. Push all pixels e such that
min [( fmax, μ(c,d) )] > f(e) to Q
End if.
4. End while
5. End.
Fig.2. Fuzzy Connectedness.
Working of Fuzzy Connectedness : FC working is
presented in fig.2, FC works based on strength of
connectedness between 2 pixels. That is hanging
togetherness – based on the similarity in spatial location
as well as in intensity.
a) Fuzzy Connectedness is the strength of
connectedness of all pixels.
4
obtained by means of fixing convex hull on lung which
presence inside the rib.
B. Proposed Approach on Rib Cage Convex Hull Fitting
This is used to find the Volume of the rib cage. Fig.3,
describes the working of Rib Cage Convex Hull Fitting.
Background subtraction & Morphological operations are
used to identify ribcage and remove scapula from it. The
identified rib will be fitted by convex hull to get its volume.
This volume will be used for performing Volume difference
Test. Input of this module is LOLA images. Then extract the
rib cage part alone from the image. In this Stage, Convex
Hull technique is used to fix polygon on the extracted Rib
cage of the image. So that Volume of the Convex Hull will
be generated.
Volume Difference Test
Volume difference test is performed on two volumes got
for the same image (V1 obtained from FC and V2 obtained
from Ribcage Convex hull for the same image). The image is
said to be pathological if the difference of the two volumes
exceeds the threshold. Fig.8 shows the volume difference test
performed on the images.
The same process repeats for all the images present in
the Lola set and declares that the image has pathology or not.
Then it enters into further refinement stage for pathology
conformance.
C. Random Forest
The Classifier is built according to the training datasets
and tested effectively. See Fig.4, Random Forest is a
collection of Decision Trees in which wide range of textural
features were given to train the classifier. Test datasets will
be passed on it to perform classification. The classified result
as two main classes such as Pathological and NonPathological class. And Pathological class is further divided
into 9 categories of common Imaging signs of lungs. GLCM
(Gray Level Co-occurrence Matrix), GLRLM (Gray Level
Run Length Matrix) & Histogram feature extraction
techniques are used. Lola dataset is used for testing and for
training various pathological images. Weka tool is used for
classification. Classification result of RF is given in table 1.
Fig.3. Ribcage Convex Hull Fitting.
Working of Rib Cage Convex Hull Fitting:
Usually background of the image has higher
intensity than the object present in the image.
Histogram is used to threshold the image.
Smoothing techniques are used for image
Enhancement.
Blobs present in the image is identified by
performing Blob detection technique. Here each and
every blobs present in the image are identified with
its details such as area, perimeter, centroid etc.
Using the Blobs information, Scapula blobs can be
removed.
By using morphological operations such as
connected components, scapula blob is removed and
rib cage alone can be obtained.
Convex Hull is a technique which is used to fix hull
or mask on the image which obtained from the
previous step.
o Ie., Convex Hull is fixed on the lung which
presents inside the Rib blobs.
Fig.7 shows Convex Hull fitted on lung image presence
inside the identified rib blobs. Second Volume information is
Fig.4. Random Forest.
Feature Extraction
This approach is used to extract/select the important
features from the images. Main aim is to avoid irrelevant
feature from the image because this may leads to
misclassification when using machine learning techniques.
For Feature Selection following techniques are used.
Gray Level Co-Occurrence Matrix (GLCM) - The
co-occurrence matrix is a statistical model that is useful in a
variety of image analysis applications, such as in biomedical,
remote sensing, industrial defect detection systems, etc. The
5
gray-level co-occurrence matrix (GLCM) is a frequency
matrix, indicates the frequency of a pair of pixels that are at
“exactly the same distance and direction of the displacement
vector”.
Gray Level Run Length Matrices (GLRLM) - A
gray level run is a set of consecutive, collinear picture points
having the same gray level value. The length of the run is the
number of picture points (pixels) in the run. The major reason
for use of these features is that the lengths of the runs reflect
the size of the texture elements. It also contains the
information about the alignment of the texture. For a given
picture, it computes a set of the gray level run length matrices
(GLRM’s) for runs with any given direction.
Histogram (HIST) - Histogram is performed to show
the number of pixels in an image at each different intensity
value found in an image. It also can be used to enhance
image.
WEKA - Waikato Environment for Knowledge
Analysis is a popular suite of machine learning software
written in Java, developed at the University of Waikato, New
Zealand. It is a collection of machine learning algorithms for
data mining tasks. The algorithms can either be applied
directly to a dataset or called from Matlab code.
Weka contains tools for data pre-processing, classification,
regression, clustering, association rules and visualization.
Fig.6. FC Segmented Lung.
Convex hull Fitted on Rib Lung Image
Fig.7.RibCage Fitted Lung Image
Weka (For Classification)
Fig.5. Common Signs of Lung diseases.
V.
EXPERIMENTAL SETUP & RESULTS
The input LOLA dataset[1] is collected from Lola
website. FC is used for segmenting the lung. Classification is
done by using WEKA tool. Train datasets are collected from
various lung databases[11] which includes signs of lung
diseases, pathological lung images & normal lung images.
LOLA dataset is used as test dataset in Classification.
FC Segmentation
Fig.8. Random Forest Classification.
Performance Measures
Comparision with Other Segmentation Technique:
FC and RF methods are considered to be best for
identifying pathological lung with signs. Table 1 holds
Comparision details of Segmentation Algorithm.
6
S.No
Segmentation
1
Threshold-based
Segmentation
Region-based
Segmentation
Fuzzy-Connectedness
Segmentation
2
3
images. Further, the robustness and the effectiveness of the
proposed method was tested on more lung scans acquired
through various sources containing a wide range of
abnormalities. And the classifier performance is compared
with other classifier. So, this confirm that the best accuracy
and effectiveness of the presented method.
Future work is to mainly concentrate on identifying
the particular type of lung diseases with its severity.
Accuracy
Percentage(%)
70%
82%
98%
Table 1 Segmentation Performance.
REFERENCES
Comparison With Other Classifiers:
Weka tool is used for Classification. Below Table.2
holds the information about various Classifiers, its time taken
and accuracy percentage obtained.
S.No
Classification
Time
Accuracy
(Sec)
Percentage(%)
1
REP
0.5
99.7809%
2
Decision
0.9
99.7809%
3
Adaboost
0.19
99.0142%
4
Bagging
0.16
99.7809%
5
RF (Proposed)
0.17
99.8905%
[1]
[2]
Table 2 Classifiers Performance.
[3]
Here the same training dataset and test dataset were applied
to various classifier to check their performance. From this
analysis, The RF(Random Forest) Classification is the one
which classifies the lung with greater accuracy with
reasonable time.(Given in fig.9).
[4]
[5]
[6]
[7]
[8]
Fig.9. Classifiers Performance Chart.
The above analysis shows that FC and RF are best to identify
signs of pathological images.
[9]
CONCLUSION AND FUTURE WORK
The proposed method is the first generic approach
covering a wide range of pathologies. The core of this work
is FC segmentation and rib cage volume estimation. This
method is equipped with multiple refinement stages including
Random Forest machine-learning classification to handle
pathologies. Also the classifier is trained with more than 900
[10]
7
Awais Mansoor, Member IEEE, Ulas Bagci, Member
IEEE, Ziyue Xu, Brent Foster, Kenneth N. Olivier,Jason
M. Elinoff, Anthony F. Suffredini, Jayaram K. Udupa,
Fellow IEEE and Daniel J. Mollura."A Generic
Approach to Pathological Lung Segmentation". IEEE
Transactions on Medical Imaging, Vol. 33, No. 12,
December 2014.
American Lung Association, Epidemiology and Statistics
Unit Research
and Program Services, “Estimated
prevalence and incidence of lung disease by Lung
Association Territory,” Lung Assoc, Washington, Tech,
2010.
G. J. Kemerink, R. J. Lamers, B. J. Pellis, H. H. Kruize,
and J. Van Engelshoven, “On segmentation of lung
parenchyma in quantitative computed tomography of
the lung,” Med. Phys., vol. 25, p. 2432, 1998.
S.G.Armato, III and W. F. Sensakovic, “Automated lung
segmentation for thoracic CT: Impact on computeraided diagnosis,” Acad. Radiol., vol. 11, no. 9, pp. 1011–
1021, 2004.
T.S. Chen, P.C. Chung, P.S. Liao, “A Fast Algorithm for
Multilevel Thresholding,” Journal of Information
Science and Engg, pp.713-727,2001.
M.N.Prasad, M.S. Brown, S. Ahmad, F. Abtin, J. Allen, I.
daCosta, H.J. Kim, M. F. McNitt-Gray and J. G. Goldin,
“Automatic segmentation of lung parenchyma in the
presence of diseases based on curvature of ribs,” Acad.
Radiol., vol. 15, no. 9, pp. 1173–1180, 2008.
Karthick S, Puraneeswari A and Sathiyasekar K, “A
Survey Based on Region Based Segmentation,”
International Journal of Engineering Trends and
Technology (IJETT) – Volume 7,2010.
M. Arfan Jaffar, Ayyaz Hussain, Anwar Majid Mirza,
“Fuzzy entropy based optimization of clusters
for the segmentation of lungs in CT scanned
images,” Knowledge Info Systems,Vol.24, pp. 91-111,
July 2010.
Timothy N.Jones, Dimitris N.Metaxas, “Image
Segmentation Based on the Integration of Pixel
Affinity and Deformable,” published in the Proceedings
of CVPR,1997.
A. Prabin, J. Veerappan, “Automatic Segmentation of
Lung CT images by CC based Region Growing,”
Journal of Theoretical and Applied Information
Technology. Vol. 68 No.1,October 2014.
[11]
[12]
Xiabi Liu, Ling Ma, Li Song, Yanfeng Zhao, Xinming
Zhao and Chunwu Zhou, “Recognizing Common CT
Imaging Signs of Lung Diseases Through a New
Feature Selection Method Based on Fisher Criterion
and Genetic Optimization,”IEEE Journal of BioMedical
and Health Informatics, Vol.19, no.2, March 2015.
Awais Mansoor, Brent Foster, Daniel J. Mollura, Kenneth
N. Olivier, Jason M. Elinoff, Jayaram K. Udupa,
Suffredini, Ulas Bagci and Ziyue Xu, “A Generic
Approach to Pathological Lung Segmentation,” IEEE
Transactions on Medical Imaging, Vol. 33, No. 12,2012.
8
Ensemble Approach on Maize Leaf Nutrition
Deficiency Identification Using Enhanced
Multivariate Partial Least Square
S. Sridevy
Dr. Anna Saro Vijendran
Assistant Professor
Department of PS and IT, AEC and RI
Tamilnadu Agricultural University
Tamilnadu, India
Director & Head
Department of MCA
SNR Sons College
Coimbatore, Tamil Nadu, India
Abstract—With the evolution of technology, people have
adopted their day today lives to utilize the benefits of
highly advanced technologies. Plants are among the
earth's most useful and beautiful products of nature.
The crucial need is that many plants are at the risk of
extinction. The proposed system implements a novel
approach to identify the lack of nutrients in the crop
image based on feature extraction. Images are collected
from the Agricultural University, Coimbatore. The
input images are processed in three steps:
preprocessing, feature extraction and regression. In the
preprocessing, HSV transformation and histogram
enhancement are processed. After the preprocessing,
features of images are extracted based on their unique
characteristics of images they are colour, shape and
texture. Based on the extracted features to identify the
nutrition deficiency in maize leaf images multivariate
partial least square is used as image analyzing
techniques. The experimental result shows the more
promising result on the proposed work while comparing
the existing approaches.
Feature extraction is a general term for methods of
constructing combinations of the variables to get
around these problems while still describing the data
with sufficient accuracy. Automatic detection of plant
diseases is an important research topic as it may
prove benefits in monitoring large fields of crops, and
thus automatically detect the diseases from the
symptoms that appear on the plant leaves. This
enables machine vision that is to provide image based
automatic inspection, process control and robot
guidance. Comparatively, visual identification is
labor intensive, less accurate. When crops suffered
disease, if the symptoms are not obvious or more
complex, it will be very difficult to distinguish the
characteristics so that delay the effective control of
crop diseases seriously. We can analyze the image of
disease leaves by using computer image processing
technology and extract the features of disease spot
according to color, texture and other characteristics
from a quantitative point of view. The caused and
extent of disease can be diagnosed timely and
effective, it could be prevented and control
comprehensive combined with the specific
circumstances of crop. It has a great significance in
automatic and intelligent management aspects of crop
growth and health with the research of crop diseases
using image feature extraction technology.
The quality inspection of leaves consists of two
main aspects, internal and external examinations.
Human sensory usually achieves the internal quality
inspection, smoking test or chemical analysis, while
the external quality inspection is mainly achieved
through human vision. It is costly and yet time
consuming to inspect internal quality since leaves
contain too many ingredients to be handled. As an
alternative, external quality examination is often used
Index Terms—image processing, agriculture, maize
crop, multivariate partial least square and feature
extraction.
I.
INTRODUCTION
Feature extraction involves simplifying the amount of
resources required to describe a large set of data
accurately. When performing analysis of complex
data one of the major problems stems from the
number of variables involved. Analysis with a large
number of variables generally requires a large
amount of memory and computation power or a
classification algorithm which over fits the training
sample and generalizes poorly to new samples.
9
instead in the examination of internal quality of
leaves, since external features are closely related to
internal quality. The external quality inspection of
leaves includes judgment of color, maturity, surface
texture, size and shape. Human vision, which is
inevitably limited by personal, physical and
environmental factors, has been the predominant
means of inspection.
Research has received considerable attention in
recent years [1, 2] on the utilization of moments for
object characterization in in-variant tasks and noninvariant tasks. The mathematical concept of
moments has been around for many years. It has been
used in various fields varying from mechanics and
statistics to pattern recognition and image
understanding. This paper focuses on automation
through computer vision. Computer vision [3] is
concerned with the theory behind artificial systems
that extract information from images. This work was
done to extracts different features from maize leaf
images based on color, texture and shape.
Furthermore, the system used the results of the
classification scheme in identifying the class of
nutrition deficiency of maize leaf images
Phosphorous deficiency using artificial neural
network segmentation and Otsu method.
Marian wiwart et al [7] adapted Euclidean
distances between the color of leaves at successive
nodes the change of color in leaves (Faba bean, pea,
yellow lupine plant) were analyzed. These color
variations in leaves was due to lack of Nitrogen,
Potassium, phosphorus and Inverted-V deficiency
respectively.
Jyotismita Chaki et al. [8] worked on a complete
plant leaf recognition system which had two stages.
The first stage was the extraction of leaf features.
Moment invariants technique was often used as
features for shape recognition and classification. The
second stage involved classification of plant leaf
images based on the derived feature obtained in the
previous stage. Comparing descriptors of the
unknown object performed classification. The
features were given as input to the neural network
which was based on supervised learning algorithm
having multilayer preceptor with feed-forward back
propagation architecture. The result of this ranged
from 90-94% accuracy having mean square error in
the range of 0-0.9.
Stephen Gang Wu et al. [9] worked using PNN
(Probabilistic Neural Network) for classification of
leaf images based on 12 leaf features. A principal
component analysis was used for reducing the 12
dimensions into five dimensions for faster
processing. The 12 features used were physiological
length, physiological width, leaf area, and leaf
perimeter, smooth factor, aspect ratio, form factor,
rectangularity, narrow factor, perimeter ratio of
diameter, perimeter ratio of physiological length and
physiological width, and vein features. If a Standard
size image was not used then the features of the same
leaf vary with different sizes images. Since, it was the
same leaf, the value of all the features were expected
to be same. The values for these features vary for
different scaled versions of the same leaf image. In
this research, features used were scaling, translation
and rotation invariant.
Y. Nam et al. [10] studied a shape-based leaf
image retrieval system using two novel approaches.
These were named as, an improved maximum power
point algorithm and a revised dynamic matching
method. In his approach, they used a region-based
shape representation technique to define the shapes of
the leaf. Certain points of interest were evaluated and
the distances between the points were taken as the
features for image matching.
Bivashrestha et al. [11] studied various features
related to the shape of the leaves for leaf image-based
II. LITERATURE SURVEY
This section provides some discussion about
several applications on feature extraction on image
processing techniques in agriculture available today.
LiliMa et al [4] in their work studied the Nutrient
deficiency found in soya bean plant by considering
six stages of growth in soya bean plant with the
application of nitrogen fertilizer that ranges between
0%-150% in each stages . The noises were removed
and the object of interest was separated between the
backgrounds using minimum error threshold method.
The color appearances of soya bean leaves were
evaluated using RGB and HIS color space models.
The extinct difference between the normal and the
leaf with variation was found using image processing
method.
R.A.DL.Pugoy et.al [5] presented in their paper
about rice, which is one of the domestic plant in
worldwide and analysis in this crop, is very
important. The various nutrients deficiency, nutrients
toxicity and the type of diseases from the leaves of
rice plant using image processing technique
comprising clustering and spatial analysis in order to
obtain the color variations on leaf were analyzed.
Maicon A. Sartin et al [6] designed image
processing system for precision agriculture to
improve the agriculture production system in cotton
plant to find macro nutrients deficiency especially
10
plant classification. The features selected in this
automated plant identification were isoperimetric
quotient, convex hull ratio, aspect ratio and
eccentricity. These features were vital in identifying
leaf shapes, number of lobes and margin types.
Leaves of nine classes were used and a classifier was
implemented and tested using 105 leaves of 38
different species. K-means clustering was used to
classify 105 leaf images into nine clusters.
R. Janani et al. [12] proposed a method for the
extraction of shape, color and texture features from
leaf images and training an artificial neural network
(ANN) classifier to identify the exact leaf class. The
key issue lies in the selection of proper image input
features to attain high efficiency with less
computational complexity. They tested the accuracy
of network with different combination of image
features. The tested result on 63 leaf images revealed
that their method gave the accuracy of 94.4% with a
minimum of eight input features. Image processing
and neural network toolboxes were used in
MATLAB to implement the work.
Zheng Niu et al [13] with four wavelengths and a
supercontinuum laser as a light source was designed
to monitor the fine structure and the biochemical
parameters of vegetation.
Pawan et al [14] in their proposal initially
preprocessing the input image using histogram
equalization is applied to increase the contrast in low
contrast image, K-means clustering algorithm is used
for segmentation which classifies objects based on a
set of features into K number of classes and finally
classification is performed using Neural-network.
Thus image processing technique is used for
detecting diseases on cotton leaves early and
accurately. It is used to analyze the cotton diseases
which will be useful to farmers.
In this paper, the salient features of leaves were
studied. Recognition was studied to develop an
approach that produced the better feature extraction
technique. This work was done on maize leaf and
extracting the colors, texture and shape features. It
was based on local region leaves and was tried to get
better accuracy results as compared to previous
works.
III. METHODOLOGY
Image
Acquisition of
Maize Leaf
Image
Preprocessing
Images
Feature Extraction
Color Feature
Extraction
Shape Feature
Extraction
Texture Feature
Extraction
Nutrient Deficiency
Identification using
Multivariate Least Partial
Square
Figure 1 Block Diagram of the Proposed Maize Leaf
nutrition deficiency Recognition System
The proposed architecture of this research work is
shown in the figure 1. The dataset used in this work
is maize leaf images according to the
recommendation given by AgriPortal of Tamilnadu
Agricultural University, Coimbatore [15]. The
images are preprocessed prior to the identification of
presence of deficiency. Initially the given input
images are converted to HSV Color Transformation
to determine the intensity and color variance near the
edges of the objects. Next Histogram equalization is
applied to produce gray map which increases the
intensities can be better distributed on the histogram.
Different features of the leaf are obtained and these
features are given as input to the Multivariate Least
Partial Square. Training and testing of the proposed
work is carried out and the result is obtained.
The features used in leaf recognition are described
as follows:
Block diagram of the proposed method is shown in
Figure1.
11
A.
Shape Features
Moment 3: It is called Skewness. It gives measure of
the degree of asymmetry in the distribution.
There are several kinds of geometric features
involved as shape features:
Aspect ratio: The ratio of maximum width to
major axis length.
Width ratio: The ratio of width at half of
major axis to maximum width.
Apex angle: The angle at apex between leaf
edges on width line at 3/4th of the major
axis.
Apex ratio: The ratio of the angle made at
the apex by width at 9/10th of major axis to
angle at 3/4th of major axis.
Base angle: The angle at base between leaf
edges on width line at 1/4th of the major
axis.
Moment ratio: The ratio between Y
deviation and X deviation from the centroid
of the leaf. It accounts for the mass
distributed around the leaf, whether it is in
longitudinal or lateral direction. If the value
is greater than one, the leaf is longitudinal in
nature; and vice versa. Whereas, a value
equal to one classifies it as circular.
Circularity: The ratio of 4*PI*Area of the
leaf to the square of perimeter.
Moment 4: It is called Kurtosis. It gives the measure
of peakedness in the distribution.
C. Texture Features
An image texture is a set of standard of
measurements computed in image processing
intended to enumerate the apparent texture of a leaf
image. Leaf image texture gives information
regarding the spatial arrangement of colour or
intensities in a leaf image or selected region of a leaf
image. Texture features are extracted from Gray
Level Co-occurrence Matrices (GLCMs). GLCM is
very useful to obtain valuable information about the
relative position of the neighbouring pixels in an
image [7]. The co-occurrence matrix GLCM (i,j)
counts the co-occurrence of pixels with gray value i
and j at given distance d. The direction of
neighbouring pixels to represents the distance can be
selected, for example 135o, 90o, 45o, or 0o, as
illustrated in Figure 2 .
B. Color Features
Color features are essential for colour based image
analysis. The information of colour distribution in an
image can be captured by the low order moments.
The first & second order moment has been proved to
be efficient and effective in representing colour
distribution of image.
 Moment 1: It is called Mean. It provides average
Color value in the image.
Where MN represents the total number of pixels in
the image
Figure 2: Directions in calculating GLCM
 Moment 2: It is called Standard Deviation. The
standard deviation is the square root of the variance
of the distribution
D. Multivariate Partial Least Square (MPLS)
Multivariate data analysis is the simultaneous
observation of more than one characteristic. In
contrast to the analysis of univariate data, in this
approach not only a single variable or the relation
between two variables can be investigated, but the
12
relations between many attributes can be considered.
In general consider an multivariate linear regression
model with an n× m matrix X of predictors and an n
× p matrix Y of dependent variables; both matrices
are mean-centered. Similar to Principal component
regression (PCR), approximate
or
In this equation T and U are the respective score
matrices that consist of linear combination of the x
and y variables. Then P and Q are the respective
loading matrices of X and Y. Note that, unlike in
PCR, in PLS in general not the loadings are
orthogonal, but the scores a ≤ min(n; m) is the
number of PLS components. Additionally, the x- and
y-scores are related by the so-called inner relation a
linear regression model maximizing the covariance
between the x- and y-scores.
The regression coefficients are stored in a diagonal
matrix and the residuals in the matrix H.
Figure 3: Flowchart of MLPS
The inner relation can destroy the uniqueness of the
decomposition of the data matrices X and Y.
Normalization constraints on the score vectors t and u
avoid this problem. To fulfill these restrictions we
need to introduce (orthogonal) weight vectors w and
c such that
The figure 3 shows the overall work flow of MLPS in
which it first generate the matrix and convert it into
linear regression, next it generate the diagonal matrix
for regression then to normalization process is
performed for finding the maximum covariance
which helps in prediction of deficiency part of the
input maize leaf image.
With
IV. EXPERIMENTAL RESULTS
Consequently, here the score vectors do not result
from projection of X on loading vectors but on
weights.
The experimental process of nutrient deficiency
identification using Multivariate Partial Least Square
(MPLS) using enhanced feature extraction has been
The objective function of PLS regression can then be
written as
deployed using MATLAB. Control and deficient leaf
images are collected from experimental fields of
Tamilnadu Agricultural University, Coimbatore [13].
Total 60 images are collected from open source, in
13
those 30 images for training purpose and 30 images
for testing purpose.
Table1: Performance Comparison of Proposed Work
with exisitng Approaches based on Precision and
Recall
Measures
Multivariate
Image
Analysis
Original Image
RGB to HSV
Multivariate
Least
Partial
Square
image
analysis
Enhanced
Multivariate
Least
Partial
Square
image
analysis
Precision
0.75
0.86
0.96
Recall
0.68
0.79
0.94
Deficiency Part
Histogram Plot
Figure 4: Shows the output the proposed work
A. Performance Evaluation Metrics for Enhanced
MLPS Systems
In EMLPS, the most commonly used performance
evaluation measures are Precision and Recall [10].
Precision is defined as the ratio of the number of
relevant images retrieved to the total number of
images retrieved.
Figure 5 :Comparison of Proposed Work with
exisitng Approaches
P = Number of relevant maize leaf images retrieved
total number of maize leaf images retrieved
The result shows that the proposed work performs
better due to the potential feature extraction
techniques based on color, shape and texture. Thus it
shows the importance and selection of appropriate
feature extraction technique for better analysis of
nutrition deficiency in maize leaf images.
R = Number of relevant maize leaf images retrieved
total number of maize leaf images in the database
The table1 shows the precision and recall value of
proposed enhanced Multivariate least partial square
image analysis with other two existing approaches
namely multivariate image analysis and traditional
multivariate least partial square image analysis.
14
With Artificial Neural Network For Nutrient
Deficiency In Cotton Crop‖. Journal Of
Computer Science 10 (6): 1084-1093.
V. CONCLUSION
Incorporating feature descriptors is a feasible
alternative for classifying structurally complex
images. They offer exceptional invariance features
and reveal enhanced performance than other moment
based solutions. They can be computed in parallel
and as the computational performance of computers
increase; the time necessary for their calculation
perhaps will not be a problem in the nearby future.
The database is prepared for the experimental use.
The database contains various leaves with various
shapes, colours and size. Experiment is carried out
with the different maize leaves of different classes
and tested. This paper investigated the use of a
number of different colours, texture and shape
features for maize leaf nutrition deficiency
identification system. The features obtained from
colour moments can be further used for indexing
images based on colour. It can be further used for
classification of leaves into respective nutrition
deficient using Multivariate Partial Least Square
image analysis.
[7] Marian
Wiwarta,
Gabriel
Fordon
ski,
KrystynaZuk-Gołaszewska.
2009.
―Early
diagnostics of macronutrient deficiencies in three
legume
species
by
color
image
analysis.Computers
and
Electronics
in
Agriculture‖ 65: 125–132.
[8]
Jyotismita Chaki, Ranjan Parekh, ―Plant Leaf
Recognition using Shape based Features and
Network classifiers‖ (IJACSA) International
Journal of Advanced Computer Science and
Applications, Vol. 2, Year 2011.
[9] Stephen Gang Wu, Forrest Sheng Bao, Eric You
Xu, Yu-Xuan Wang, Yi-Fan Chang and QiaoLiang Xiang, ―A Leaf Recognition Algorithm for
Plant Classification Using Probabilistic Neural
Network‖,arXiv:0707.4289v1 [cs.AI], Year 2007.
[10] Y. Nam, ―Elis: An efficient leaf image retrieval
system‖, in Proc. Advances in Pattern
Recognition Int. Conf., Kolkata, India. Year 2005
[11] Bivashrestha, ―Classification of plants using
images of their leaves‖ ICPR, vol. 2, pp. 25072510, Year 2000.
REFERENCES
[12] R. Janani, A. Gopal, ―Identification of selected
medicinal plant leaves using image features and
ANN‖, Advanced Electronic Systems (ICAES),
2013 International Conference, Page(s): 238 –
242, Year 2013.
[1] M K HU, ―Visual pattern recognition by moment
invariants‖, IRE Trans. Info Theory, vol. 8, pp.
179-187, Year. 1962.
[2] T. H. Reiss ,―The Revised Fundamental Theorem
of Moment Invariants‖, IEEE Transaction Pattern
Anal. Machine Intelligence, Vol. 13, No. 8, Page
830-834, Year. 1991
[13] Zheng Niu, Zhigang Xu, Gang Sun, Wenjiang
Huang, Li Wang, Mingbo Feng, Wang Li,
Wenbin He, and Shuai Gao, ―Design of a New
Multispectral Waveform LiDAR Instrument to
Monitor Vegetation‖, IEEE Geoscience And
Remote Sensing Letters, vol. 12, no. 7, july 2015
[3] J. Du, D. Huang, X. Wang, and X. Gu,
―Computer-aided plant species identification
(CAPSI) based on leaf shape matching
technique,‖ Transactions of the Institute of
Measurement and Control. Vol. 28, No.3 Page.
275-284, Year 2006.
[14] Pawan P. Warne, Dr. S. R. Ganorkar, ―Detection
of Diseases on Cotton Leaves Using K-Mean
Clustering,‖ International Research Journal of
Engineering and Technology (IRJET), Volume:
02 Issue: 04 , July-2015
[4] Lili Ma, Junlong Fang, Yuehua Chen, Shuipeng
Gong. 2010. ―Color Analysis of Leaf Images Of
Deficiencies And Excess Nitrogen Content In
Soybean Leaves‖. E-Product E-Service And EEntertainment (ICEEE), Vol., No., 1-9.
[15] URL: http://agritech.tnau.ac.in
[5] PugoyR.A.Dl.and
V.Y.Mariano,
2010.
―Automated Rice Nutrient Deficiency, Nutrient
Toxicity And Disease Detection Using Digital
Image Analysis‖. 40th Cssp Anniversary And
Annual Scientific Conference, Davao City, 1520.
[6] Maicon A. Sartin, Alexandre C.R. Da Silva and
ClaudineiKappes, 2014. ―Image Segmentation
15
Low Power Multiplier Using Algorithmic Noise
Tolerant Architecture
S.Fathima
Dr.S.Uma
PG scholar, Dept. of ECE
Coimbatore Institute of Technology
Coimbatore, India.
Fathimait2014@gmail.com
Associate Professor, Dept. of ECE
Coimbatore Institute of Technology
Coimbatore, India.
Uma@cit.edu.in
Abstract—In this paper, Fixed-width Reduced Precision
Replica Redundancy Block (RPR) design is adopted in
Algorithmic Noise Tolerant (ANT) architecture. The main
objective of this design is to achieve High Precision, Low power
and high speed multiplier. By using Fixed- width RPR
computation error can be identified and corrected. The ANT
architecture was implemented using Xilinx and ModelSim tool.
Performance metrics of power is compared with existing Fullwidth RPR designs. Average Power consumption of the proposed
design is found to be 74.42µW at 100Hz. Proposed design achieves
less power and area efficiency than the Full-width RPR.
II. DESIGN METHODOLOGY OF ANT ARCHITECTURE USING
FIXED-WIDTH RPR.
Unlike Full-width RPR which increases the bit width, Fixedwidth RPR avoids infinite growth of bit-width. ANT
architecture as shown in Fig 1, includes Main Digital Signal
Processor (MDSP) and Error Correction (EC) block [1]. To
reduce the power consumption in Main block VOS is used. But
in VOS whenever critical path delay Tcp is greater than
sampling period Tsamp, soft error occurs which leads to severe
degradation of incoming signals. EC block contains reduced
computation delay. Under VOS, there are number of inputs
x[n] and its corresponding output of Main block is ya[n] and
output of Fixed-width RPR is yr[n]. Error detection is carried
out by comparing the difference between |ya[n]-yr[n]| with the
threshold value Th.
Index Terms—ANT architecture, Fixed-width RPR, Fullwidth RPR.
I. INTRODUCTION
Rapid growth of wireless devices in recent years leads to
need for ultra low power systems. To reduce the power
consumption, low power technique such as supply voltage
scaling [1] can be used, However, Noise interference problem
arises. Hence the design technique is to be enhanced to avoid
noise interferences [2].
Final output [n] = ya[n], if |ya[n]-yr[n]| ≤ Th
yr[n], if |ya[n]-yr[n]| ≥ Th
[1]
Threshold is calculated using the formula
Th =
[2]
Where yo[n] is error free output signal. By using this technique
power can be minimized without any signal degradation.
Multipliers are key components of many high performance
systems such as FIR filters, microprocessors, digital signal
processors, etc. A system’s performance is generally
determined by the performance of the multiplier[3] because the
multiplier is generally the slowest element in the system.
Low power techniques such as voltage overscaling[4] has
been proposed to lower the supply voltage without affecting
the throughput. But it leads to more degradation in Signal to
Noise Ratio (SNR). ANT technique [2] uses VOS with RPR to
reduce the soft errors and achieves high energy. RPR design in
ANT architecture [5] is not easily adopted and repeated.
However adopting with RPR increases the area and power
consumption.
To overcome these problems, Full-width RPR is replaced
with Fixed-width RPR in the proposed method. [9] By using
Fixed-width RPR computation error can be identified and
corrected with low power consumption and less area.
Fig 1. Proposed ANT architecture with the Fixed-width
RPR[1].
This paper is presented as follows: Section II focuses on
design methodology of ANT architecture using Fixed-width
RPR. Section III elucidates the simulation of Fixed-width
RPR. Section IV narrates about the simulation results and the
comparisons and concluded in Section V.
Fixed-width RPR is applied in DSP applications to avoid the
size of growing bit-width. In Fixed-width DSP neglecting LSB
output is a popular solution. The hardware complexity and
16
power consumption of Fixed-width RPR is 50% lesser than the
Full-width RPR[6]. However, neglecting LSB bits results in
rounding error.
III. SIMULATION AND RESULTS OF FULL-WIDTH AND FIXEDWIDTH RPR
A. Error Compensation for Fixed-width RPR:
The Function of RPR is to identify and correct the error
which occurs in the output of MDSP and maintain the signal to
noise ratio by reducing the supply voltage[5]. By using Fixedwidth RPR, computation speed can be accelerated as compared
with Full-width RPR.
Fig 3. Output Wavefom of Full-width RPR.
Fig 3 explains about the simulation of Full-width RPR which
contain 12 bit with the threshold value of 10 bit. If the critical
path delay exceeds over 10 produces an error, otherwise
normal operation takes place without any error.
Fig 2. 12 x 12 bit ANT multiplier is implemented with the sixbit Fixed-width RPR.
In MDSP of 12 bit ANT Baugh-Wooley array multiplier.
That too unsigned n bit inputs of x and y is given by,
X=
xi . 2i ,
Y=
yj . 2j
[3]
The multiplied output P is the summation of partial
products of xi . yj which is given in the equation,
P=
P k . 2k =
xi yj . 2i+j
[4]
Fig 4. Gatecount of Full-width RPR.
Full-width Baugh-Wooley divided into four parts. They are
Most Significant Part (MSP), Input Correction Vector (ICV),
Minor Input Correction Vector (MICV), Least Significant Part
(LSP) as shown in Fig 3.
Fig 4 explains about the area occupied by the Full-width RPR
is 8,571 total equivalent gate count for design.
[5]
Pt =
yj 2j
xi 2i + f(EC)
[6]
Where f(EC) is the error compensation function which is equal
to the summation of f(ICV) and f(MICV).
Most of the Errors are produced at Least Significant Part, so
neglecting the LSP leads to reduction of area, which further
reduces power consumption and hardware complexity. If the
power consumption is lowered, the SNR[6] can be maintained
without severe degradation.
Fig 5. Power Analysis of Full-width RPR
17
Fig 5 explains about the power consumption of Full-width
RPR. The power achieved at 50 MHz clock frequency is 50.97
mW.
Fig 8. Power Analysis of Fixed-width RPR
Fig 8 explains about the power consumption of Fixed-width
RPR. The power achieved at 50 MHz clock frequency is 49.51
mW.
Table I
Comparison of Power Analysis of Full-width and Fixedwidth RPR
Clock
Frequency(in
MHz)
50
100
150
200
Fig 6 Output Wavefom of Fixed-width RPR.
Fig 6 explains about the simulation of Fixed-width RPR which
contain 12 bit with the threshold value of 10 bit. If the critical
path delay exceeds over 10 produces an error, otherwise
normal operation takes place without any error.
Full-width
RPR(Power in
mW)
50.97
77.35
103.72
130.1
Fixed-width
RPR(Power in
mW)
49.51
74.42
99.33
124.24
By analyzing the Table 1, Proposed method consumes less
power than existing for various frequencies.
Fig 7. Gatecount of Fixed-width RPR.
Fig 9. Bar chart of Full-width and Fixed-width RPR at various
frequencies.
Fig 7 explains about the area occupied by the Fixed-width RPR
is 7,716 total equivalent gate count for design.
Fig 9. Shows that for various clock frequencies, the power
achieved by Fixed-width RPR is Lesser than the Full-width
RPR.
At the clock frequency of 50 MHZ, the power consumed by
the full width RPR is 50.97mW whereas 49.51mW in Fixedwidth RPR for the same 50MHZ clock frequency. While
comparing with the various frequencies of clock, the power
also get reduces.
18
IV. CONCLUSION
The proposed 12 bit ANT multiplier circuit using Fixedwidth RPR is implemented using Xilinx ISE 8.1 under 1v
supply voltage and operating at the frequency of 50, 100, 150,
200 MHz respectively. The power consumption achieved at 50
MHz is 49.51 mW.
.In comparison with Full-width RPR, the circuit area and
the Power consumption is drastically reduced. This type of
ANT architecture has higher computation precision, higher
SNR and can also be operated at low supply voltage.
REFERENCES
[1] V. Gupta, D. Mohapatra, A. Raghunathan, and K. Roy,
“Low-power digital signal processing using approximate
adders,” IEEE Trans. Comput. Added Des. Integr. Circuits
Syst., vol. 32, no. 1, pp. 124–137, Jan. 2013.
[2] G. Karakonstantis, D. Mohapatra, and K. Roy, “Logic and
memory design based on unequal error protection for
voltage-scalable, robust and adaptive DSP systems,” J.
Signal Process. Syst., vol. 68, no. 3, pp. 415–431, 2012.
[3] P. N. Whatmough, S. Das, D. M. Bull, and I. Darwazeh,
“Circuit-level timing error tolerance for low-power DSP
filters and transforms,” IEEE Trans. Very Large Scale
Integr. (VLSI) Syst., vol. 21, no. 6, pp. 12–18, Feb. 2012.
[4] J. N. Chen and J. H. Hu, “Energy-efficient digital signal
processing via voltage-overscaling-based residue number
system,” IEEE Trans. Very Large Scale Integr. (VLSI)
Syst., vol. 21, no. 7, pp. 1322–1332, Jul. 2013.
[5] N. Petra, D. D. Caro, V. Garofalo, N. Napoli, and A. G. M.
Strollo, “Truncated binary multipliers with variable
correction and minimum mean square error,” IEEE Trans.
Circuits Syst., Vol. 57, No. 6, pp. 1312–1325, Jun. 2010.
[6] S. J. Jou and H. H. Wang, “Fixed-width multiplier for DSP
application,” in Proc. IEEE Int. Symp. Comput. Des., Vol.
23, No.5, pp. 318–322, Sep. 2010.
[7] I. C. Wey and C. C. Wang, “Low-error and area-efficient
fixedwidth multiplier by using minor input correction
vector,” in Proc. IEEE Int. Conf. Electron. Inf. Eng.,
Kyoto, Japan, Vol. 1, pp. 118–122, Aug. 2010.
[8] Y. Pu, J. P. de Gyvez, H. Corporaal, and Y. Ha, “An ultra
low energy/frame multi-standard JPEG co-processor in 65nm CMOS with sub/near threshold power supply,” IEEE J.
Solid State Circuits, Vol. 45, No. 3, pp. 668–680, Mar.
2010.
[9] Y. Liu, T. Zhang, and K. K. Parhi, “Computation error
analysis in digital signal processing systems with
overscaled supply voltage,” IEEE Trans. Very Large Scale
Integr. (VLSI) Syst., Vol. 18, No. 4, pp. 517–526, Apr.
2010.
19
A Survey on Compression in Encrypted Domain
T.Suguna
Dr.R.ShanmugaLakshmi
Department of Information Technology
Government College of Technology
Coimbatore, India
tsugunait@gct.ac.in
Department of Computer Science and Engineering
Government College of Technology
Coimbatore, India
drshanmi@gct.ac.in
First the sender Alice compresses the image I and
encrypts the compressed image using an encryption function
with secret key Ek(Ic). The resultant image is transferred to the
channel provider Charlie for transmitting this to receiver.
Channel provider Charlie act as a forwarder and forward the
compressed encrypted image to the receiver. Then the receiver
Bob performs decryption and decompression sequentially to
get back the original image send by the sender.
Eventhough this communication provides the secured
transmission of image, processing is fully done in the sender
side. No processing is required by the channel forwarder. In
some situation sender may be a resource restricted device. It
has limited processing capability. And also sender
concentrates only on the security of the image and not bothers
about the efficient transmission of the image. In that scenario
some compression processing is handed over to the channel
provider. For this the order of applying encryption and
compression should be changed. This system is called as
Encryption Then Compression (ETC) system. This is shown in
the following fig. 2.
Abstract—In Multimedia communication, there is a need of
large data to be transmitted through the communication channel.
To provide better efficiency of the channel, we need compression
scheme to reduce the quantity of information to be sent. In
addition to efficient transmission there is a need for secure
transmission also. To provide both security and data compaction
we need both compression and encryption. In traditional method
first compressions is applied and then apply the encryption
technique on the compressed domain. In some situation like
resource constraint environment, user needs more attention to
security of their data than compression. In this situation we need
a reverse system of the traditional compression then encryption
(CTE) system called as encryption then compression system
(ETC). In recent years a number of different methods were
proposed in this area. This paper discuss about the different
methods used for ETC system .
Index Terms—Compression,
encrypted images
encryption,
compression
of
I. INTRODUCTION
Consider a multimedia communication scenario, in which a
content owner Alice wants to securely and efficiently transmit
an image I to a recipient Bob through the untrusted
communication channel provider Charlie. Traditionally this
could be done by as shown in fig. 1
Fig. 2. Encryption Then Compression system
In this ETC system Alice only encrypt his image
using the secret key and it is transferred to the channel
provider Charlie. He was not aware of the key, so he
compresses the encrypted image. Compressed image is then
forwarded to the receiver Bob. He performs decompression
and decryption to retrieve the original image.
Fig. 1. Compression Then Encryption system
20
II. METHODS USED FOR ETC SYSTEM
Predictive encoding and encryption
For predictive encoding both non-linear and linear
predictors are available. Compared to linear predictor
suggested in JPEG spatial, non-linear predictor suggested in
JPEG-LS provides better performance for most of the natural
images. Hence in this method non-linear prediction is used.
For prediction four adjacent pixels are used. Encryption is
done in the prediction domain by performing the following
operation.
Ci=∆i+Ki
A. On Compressing Encrypted Data
Johnson et. al [2] focuses both lossless compression of
encrypted bi-level images and lossy compression of encrypted
real valued data’s separately. Distributed source coding theory
is used in this work.
Lossless Compression of Encrypted Bi-level Images
Image is considered as strings of 0’s and 1’s. The string is
encrypted by adding a unique Bernoulli string of the
appropriate length. The encrypted string is compressed by
finding its syndrome with respect to Low Density Parity
Check codes (LDPC). In the receiver side decoding and
decompression is done simultaneously using iterative
decoding algorithm specified by slepian-wolf method [1].
where, ∆i- prediction error
Ki- key
+ - XOR operation.
Decryption is performed by performing XOR operation with
cipher text and the key. Same key is maintained in both the
sides by sending random seed as side information to generate
the key for decryption.
Lossy Compression of Encrypted Real-Valued Data
The data was encrypted by a stream cipher. Encryption is
done by adding Gaussian key sequence with the data sequence
on a sample by sample basis. The encrypted data has been
compressed by the scalar quantizer and trellis code to the rate
of 1 bit/sample. Decoder has the knowledge of the key
sequence and trellis decoder. Using these two estimates
decoder compute the encrypted sequence. Form this key
sequence is subtracted to form the optimal sequence of the
original data. Encoder and decoder performance is purely
depends on the source not on the side information (key). It
achieves Distortion rate of 3-3.8 dB and Bit error rate of 10-3
to 10-4
Compression and decompression
DSC coding with irregular LDPC codes are used for
compression. This considers both gray scale and color images
for their experimental analysis. It provides lossless
compression of encrypted images. Color image is separated
into 3 color planes and apply the procedure which is applied
for grey scale image. Compression ratio achieved using this
scheme is comparatively higher than the compression of
encrypted image directly.
D. Lossless compression of encrypted grey-level and color
images
Spatial and cross plane correlation between pixels and
Correlation between color bands are used to compress the
encrypted grey-level and color images [5].
B. On compression of encrypted images
This work [3] follows the graphical model for compression
of encrypted images. This graphical model consists of three
components connected together. They are source model,
encryption model and compression model. Source model is a
2-D spatially correlated model. Which includes a image pixel
and its corresponding four neighboring pixels such as up and
down, left and right. Second model is the encryption model.
Stream cipher technique is used for encryption. Encrypted data
is obtained by performing exclusive or operation with key and
data. For compression it uses the code model, which includes
linear transformation matrix. This matrix is used for the
conversion of the encrypted data into compressed form.
Decoding is achieved by applying sum-product algorithm
on the graph model. This algorithm is an inference algorithm
and it is applied iteratively to updates an estimate of the
distribution for each of the variables. It compresses an
encrypted 10000 bits bi-level image into 4299 bits and
recovers the image exactly. It is better than the models uses
the 1-D sources.
Grey level image
Subdivide the image into bit-planes and consider each of
them as a separate black and white image. Before performing
encryption the image is stored, in such a way that spatial
correlation between bit-planes is removed. Two methods are
used to remove spatial correlation between bit-planes. First
method is a fast and simple method, in which each row is
scanned from left to right and perform XOR operation
between current bit and previous bit. Using this method
vertical edges are extracted.
Second method is very effective method applied on the
grey level values. In this method also each row is scanned
from left to right but instead of XOR operation, prediction
method is used with average of four adjacent bits to predict the
value. Then the pixel values are replaced by prediction error.
The image with the prediction error is then divided into bitplanes. To decide compression rate conditional entropy
between adjacent bit-planes is used. Probabilities are given as
the input to perform decoding and decryption.
C. Distributed Source Coding based Encryption and
lossslessCompression of Gray Scale and Color Images
Encryption is done on the prediction error domain instead
of the original image and compression is achieved by
distributed source coding [4].
Color Image
In color image each band is treated as gray scale
image. To get better compression rate, correlation between
21
reference band and other bands are used. Green color band is
selected as reference band and it is coded as grey level image.
In another approach RGB color space to YCbCr color space.
Reversible integer approximation of the correct transform is
used to attain lossless nature of the scheme while transforming
from RGB to YCbCr color space. Both XOR based and
prediction based spatial de-correlation can be applied to the
modified YCbCr and each band now treated separately as a
grey scale image for further processing.
Encoding, encryption and decoding follow the
method that is used by Johnson et al.[2] with small variation.
In encoding LDPC code is characterized by matrix having
number of rows is length of the bit planes and number
columns is length of the syndrome. Before to encryption
conditional probabilities of 0s and 1s are calculated with
respect to their reference plane and it is transmitted with the
encrypted bit-planes. Encrypted creates a file containing
header and description blocks to decide compression rate.
Decoding and decryption performed jointly. Decoder uses BP
algorithm. Each received bit plane is decoded with reference
plane and conditional probability. Each decoded bit-plane is
decrypted immediately to decode the subsequent bit-planes.
Compression
Compression of encrypted image is done with the
help of the measurement matrix Φ of size MxN. Where M<N
and that decides the compression ratio.
Y = ΦXenc
In addition to compression, compressive sensing is provide a
second level of security to the image. After compression Y
will be quantized using Lloyd quantizer to obtain Yq. This
will be send to the decoder.
Joint decoding and decryption:
Keys for encryption and compression are received via
secure channel and Yq is received via normal channel by the
decoder. Using these three values joint decoding and
decryption is done to obtain the original image.
F. Lossy Compression and Iterative Reconstruction for
Encrypted Image
In this paper pseudo random permutation based encryption
method is used to obtain the cipher text [7]. Cipher text is
compressed by removing the excessive rough and fine
information available in the transform domain. Then combined
decoding and decryption is performed by the spatial
correlation exist in the natural image. This is refined by the
iterative reconstruction process. The reconstruction process is
shown in fig.4
E. Lossy Compression of Encrypted Image by Compressive
Sensing Technique
Compressive sensing itself is provides better security for
user data. Encryption is performed in the spatial domain with
linear operations [6]. Entire process is as shown in the fig.3
Fig.3 ETC using compressive sensing
Encryption and compression is done by generating
two random matrix called as encryption matrix and
measurement matrix and the seed value for that is
communicated to the decoder through the secure channel.
Using this, the decoder decrypts and decodes the received
information.
Fig. 4 Reconstruction process
G. On Compression of Data Encrypted With Block Ciphers
Usually encryption is done in two ways. It may be stream
cipher or block cipher. In stream cipher sequence of bits are
considered for encryption using XOR operation. But in case of
block ciphers block of codes are used. AES (Advanced
Encryption Standard) and DES(Data Encryption Standard) are
examples for block cipher encryption. In block ciphers there is
practical limitation of mapping two separate plain text block to
a single cipher block. So in this method block cipher with
chaining mode like CBC(cipher block chaining) and Output
Feedback (OFB) or Cipher Feedback modes are used. Post
Encryption
Cipher text Xenc is generated by converting the given
image MxN into single dimensional matrix X of size N and
then apply the linear transformation. The process of
encryption is shown by the following relation.
Xenc = μX
Where μ is the matrix of size NxN.
22
Encryption Scheme (PES) based on the Slepian-Wolf coding
is used with block cipher chaining modes to perform
encryption.
Compression is done by exploiting the correlation
between the plain text and cipher text without having any
knowledge about the key. Initialization vector and cipher text
blocks are given as the input to the compressor. Compressor
applies Slepian-Wolf encoding procedure to compress the
input except the last cipher block. Block diagram of the
compresser is shown if fig.5
Compression
Compression of encrypted images is done by adaptive
Arithmetic Coding. The channel provider applies the adaptive
arithmetic coding for the received prediction error sequence
and generate binary bit streams. Length of the bit stream is
sent as side information for decryption and decompression.
Decompression and decryption:
Decompression and decryption is performed jointly at the
receiver side using the side information. Receiver divides the
received bit streams into L clusters then applies adaptive
arithmetic decoding and de permutation to get back the
original cluster. With all the cluster values, decoding of each
pixel is done in raster scan order. Bytes and bits per pixel are
used as performance measure. Very small (<0.2%) amount of
coding penalty is incurred.
J. Designing
an
Efficient Image
Encryption-ThenCompression System via Prediction Error Clustering and
Random Permutation
Image encryption is performed on prediction error domain
and random permutation [11]. Adaptive arithmetic coding is
used to losslessly compress the encrypted image.
Fig. 5 Block diagram
Receiver uses joint decompression/decryption
method to get the original information. First Slepian-Wolf
decoding and decryption procedure is applied from right side
block that is from last block to the left block that is
initialization vector to get the original information.
Image encryption
The schematic diagram of the image encryption is as shown
in the fig. 6
H. Compression of encrypted images with multilayer
decomposition
Multi layer decomposition [9] is used for lossy compression
of encrypted gray scale images.
Encryption operation is starts from dividing images into sub
images. Prediction errors are calculated for different layers of
the image. Then the sub image and prediction errors are
encrypted by exclusive-or operation and a pseudo random
permutation. Channel provider compresses the encrypted
images by performing quantization and optimization with rate
distortion criteria on various layers of the permuted prediction
errors. At the receiver, compressed encrypted image is
dequantized and decrypted with the help of the key known by
the receiver.
Fig.6 Image Encryption
Input image is given to the GAP predictor, which predicts
the error. Instead of encoding original pixels, here prediction
errors are encoded. To encode this, the prediction errors are
mapped to one of the values [0-255]. Then these mapped
values are divided into L number of clusters and apply the
random permutation to obtain the encrypted image.
I. On the design of an efficient encryption-then-compression
system
Encryption
In this paper image encryption is done by prediction error
clustering and random permutation [10]. Prediction of each
pixel is made by using image predictor GAP because of its
excellent de-correlation capability. Predicted value Īi,j of each
pixel Ii,j is refined by feedback mechanism to obtain correct
prediction value Ĩ i,j. prediction error associated with each pixel
is calculated by
e i,j= Ii,j - Ĩ i,j
Instead of applying encryption in the prediction domain as a
whole, it divides the prediction errors in to L clusters. Then
each cluster is reshaped and permuted by cyclic shifts. Finally
encrypted image is generated by concatenating all the
permuted clusters.
Lossless compression
Channel provider does not know the encryption key used by
the sender. So to compress the encrypted image cluster
information is passed as the side information. Using this
compression takes place adaptive arithmetic coding. First
predicted error sequence is converted into binary bit streams
parallel to improve the performance. Binary bit streams are
combined to form bit sequence. The length of the bit stream is
sent as side information to the receiver Bob.
23
Sequential Decompression and Decryption
Upon receiving compressed encrypted bit stream, for each
bit stream Bob applies adaptive arithmetic decoding to obtain
permuted predicted bit stream. Then this was de permuted by
the secret key known by the receiver. To obtain the original
reconstructed image raster scan method is used.
quantized and then are encoded by embedded block coding
with EBCOT followed by rate control operations. Rate control
is used to make the bit stream conform to a target size. Finally,
a JPEG 2000 compliant bit stream is generated by adding
packet headers, a main header and other control codes.
Perceptual encryption makes an image difficult to
understand visually. The feature similarity (FSIM) index
between original image I and encrypted image Ie was
calculated to confirm perceptual encryption. The average Peak
Signal to Noise Ratio (PSNR) comparison is done to ensure
efficiency in compression. Receiver performs decompression
and decryption to obtain the original image.
K. A game theory-based block image compression method in
encryption domain
This paper proposes block based compression technique in
encryption domain to provide better security as well as
compatibility to the service providers. Block based
compression is formulated as a game theoretical problem and
it is optimized by game theory [12].
Encryption of the image is done by two steps. First entire
image is divided into blocks and perform the block
permutation. Second step is within each block perform
position permutation. Key derivation function is used to
generate the keys requires to perform inter block and intra
block permutation. Same key generation mechanism is used
by both encoder and decoder. Only seed for key generation is
transmitted to the decoder to generate the required keys.
Decryption is also very similar to the encryption process.
Compression of encrypted image is done by forming game
theoretical problem in the game theory. Game theory is used
to investigate the trade-off between quality of the image and
desired number of bits used to represent it. Each encrypted
block has a player to compete in the game. The strategy of the
player is to control the number of bits used to represent the
compressed stream in such a way that to provide good quality
reconstructed image.
For decompression and decryption along with the
compressed encrypted data, seed of the key is also delivered.
Receiver uses this information to reconstruct the original
image with good quality.
M. An encryption-then-compression system for JPEG
standard
Encryption is done using block based perceptual
encryption that is compatible to use JPEG compression for
encrypted image [14].
Encryption
Divide each color component of a color image
I={IR,IG,IB} into Bx* By blocks respectively. Permute
randomly the divided blocks using a random integer generated
by a secret key K1 (Block Scrambling). Rotate and invert
randomly each block using a random integer generated by a
key K2 (Block Rotation and Inversion). Apply the negativepositive transformation to each block using a random integer
generated by a key K3 (Negative-Positive Transformation).
Shuffle three color components in each block using a random
integer generated by a key K4 (Color component Shuffling).
After applying these four steps, generate the encrypted image
by integrating the transformed block images. The entire
process is shown in the following fig. 7
L. An Encryption-then-Compression system for JPEG 2000
standard
In this paper encryption and compression is based on
the Discrete Wavelet Transformation co-efficient [13].
Encryption
For encryption, the JPEG 2000 compliant DWT is applied to
Image I. Then, the DWT coefficients, C, which are X *Y in
size, are obtained. DWT coefficients C are perceptually
encrypted into Ce with two types of encryption schemes
namely Sign Scrambling and Block Shuffling based on a
pseudo-random number generator (PRNG) with a secret key.
Encrypted DWT coefficients Ce are transformed to a spatial
image by using inverse DWT. The spatial image is linear
quantized to maintain the signal range of the input image, and
then perceptually encrypted image Ie is obtained.
Fig.7 Four steps of encryption
Compression
Compression is done for both gray scale as well as color
images. To encode gray scale image, four basic steps are
followed. They are:
Dividing an image into non-overlapped
consecutive 8* 8 blocks.
Applying 2-D DCT to each block.
Block-based quantizing.
Entropy coding using Huffman coding.
Compression
Encrypted image is then compressed by channel provider. For
performing compression, the subbands are divided into cx *cy
sized code-blocks. DWT coefficients in the code-blocks are
24
Quality of an encrypted and compressed image is almost as
same as that of compressed original image and an encrypted
image has almost the same image quality as another encrypted
image, although they have different visibility. Large Key
Space allows for the increased security of the images.
On the other hand, in the JPEG compression of a color image,
two processing steps are added before the steps of gray scale
image compression. They are
Performing a color- space transformation from RGB
to YCbCr.
Sub-sampling the Cb and Cr components to reduce
the spatial resolutions.
Performing the above steps to the brightness component Y and
the sub-sampled chroma components independently and
combine the result to obtain the compressed image. Image
III. COMPARISON OF DIFFERENT METHODS
Following table Tab.1 shows the comparison different
methods used for compression and encryption with test
images and performance metrics.
Table.1 Comparison Table
Sl.No
1.
2.
3.
4.
5.
6.
7.
Title
On compressing encrypted
data [2]
On compression of encrypted
images[3]
Distributed source coding
based encryption and lossless
compression of gray scale
and color images[4]
Lossless compression of
encrypted gray level and
color images[5]
Lossy
compression
of
encrypted
image
by
compressive
sensing
technique[6]
Lossy compression and
iterative reconstruction for
encrypted image[7]
Compression of encrypted
images with multi-layer
Decomposition[9]
Technique used for
Compression
Distributed source coding
Trellis code
LDPC codes
Distributed source coding
Technique used for
Encryption
Stream cipher
Performance Metrics
Encryption
model
with
stream cipher
Encryption on prediction
error (XOR)
Bi level image
Compression ratio
Gray scale
Color image
YUV color space
Gray scale
Color image
YUV color space
Lena,
Baboon, Peppers and
House, each of size
256×256
Bi level image
XOR with Bernoulli key
Bit /rate
Compressive sensing with
Gaussian
random
measurement
matrix
generated by random
matrix
Pseudo
random
permutation
Permuted
transformation
domain
PSNR
Compression ratio
Orthogonal transform
PSNR
Compression ratio
Lena
Quantization
optimization
exclusive-or operation and a
pseudorandom
permutation on permutation
errors
Prediction Error Clustering
and Random Permutation
PSNR
Compression rate
Lena, Man, Couple
and Lake
PSNR
Lena
Compression ratio
PSNR
Lena
PSNR
Bit rate
Gray scale and color
images
PSNR
Images
from
standard evaluation
material (StEM)
and
arithmetic
Linear
spatial
Designing an Efficient Image
Encryption-ThenCompression System via
Prediction Error Clustering
and
Random
Permutation[10]
Adaptive
coding
9.
A game theory-based block
image compression method
in encryption domain[12]
An
encryption-thencompression system for
JPEG standard[13]
An
Encryption-thenCompression system for
JPEG 2000 standard[14]
Game theory based bit
allocation on image blocks
Block based encryption
JPEG std
Block based
encryption
JPEG 2000
Sign scrambling and block
shuffling of DWT
11.
Bit rate
Entropy
Compression ratio
Multi level LDPC with BP
algorithm
8.
10.
Test image
25
four
step
at the cost of security. There is always a tradeoff between
security and compression efficiency in the compression of
encrypted images. Hence it is always an open issue to provide
better algorithm for compression of secured images.
Compression ratio, PSNR, Bit rate and Entropy are used as
performance metric for compression of images. Especially
compression ratio and entropy are used for lossless
compression. Using compression ratio as a performance metric
[3] and [4] were compared. Both methods considered here uses
lossless compression of encrypted images. Nearly both
methods achieved approximately same compression ratio in the
range of 2 to 2.5. Encryption used in [3] and [4] is simple
stream cipher with XOR operation. Even encryption used here
is simple method, it provides better security using enough key
space.
REFERENCES
[1] D. Slepian and J. K. Wolf, “Noiseless coding of correlated
information sources,” IEEE Trans. Inform. Theory, vol. IT-19,
pp. 471–480, July 1973.
[2] M. Johnson, P. Ishwar, V. M. Prabhakaran, D. Schonberg, and
K. Ramchandran, “On compressing encrypted data,” IEEE
Trans. Signal Process., vol. 52, no. 10, pp. 2992–3006, Oct.
2004.
[3] D. Schonberg, S. C. Draper, and K. Ramchandran, “On
compression of encrypted images,” in Proc. IEEE Int. Conf.
Image Process., Oct. 2006, pp. 269–272.
[4] A. Kumar and A. Makur, “Distributed source coding based
encryption and lossless compression of gray scale and color
images,” in Proc.MMSP, 2008, pp. 760–764.
[5] R. Lazzeretti and M. Barni, “Lossless compression of encrypted
greylevel and color images,” in Proc. 16th Eur. Signal Process.
Conf., Aug. 2008, pp. 1–5.
[6] Kumar and A. Makur, “Lossy compression of encrypted image
by compressive sensing technique,” in Proc. IEEE Region 10
Conf. TENCON, Jan. 2009, pp. 1–6.
[7] X. Zhang, “Lossy compression and iterative reconstruction for
encrypted image,” IEEE Trans. Inf. Forensics Security, vol. 6,
no. 1, pp. 53–58, Mar. 2011.
[8] D. Klinc, C. Hazay, A. Jagmohan, H. Krawczyk, and T. Rabin,
“On compression of data encrypted with block ciphers,” IEEE
Trans. Inf.Theory, vol. 58, no. 11, pp. 6989–7001, Nov. 2012
[9] X. Zhang, G. Sun, L. Shen, and C. Qin, “Compression of
encrypted images with multilayer decomposition,” Multimed.
Tools Appl., vol. 78, no. 3, pp. 1–13, Feb. 2013.
[10] Jiantao Zhou, Xianming Liu and and Oscar C. Au, “ On the
Design of an efficient Encryption then Compression system, ”
ICASSP, IEEE, pp. 2872-2876, 2013.
[11] Jiantao Zhou, Xianming Liu, and Yuan Yan Tang,” Designing
an Efficient Image Encryption-Then-Compression System via
Prediction Error Clustering and Random Permutation”, IEEE
Trans. Info Forensics, vol. 9, No.1, pp. 39-50, Jan 2014.
[12] Shaohui Liu, Anand Paul, Guochao Zhang and Gwanggil Jeon,
“A game theory-based block image compression method in
encryption domain,” in springer Science, Business media New
York, April 2015.
[13] Kenta Kurihara, Sayaka Shiota and Hitoshi Kiya, “An
encryption-then-compression system for JPEG standard” in
IEEE Picture Coding symposium (PCS) , pp. 119-123, 2015.
[14] Watanab. O, Uchida. A , Fukuhara. T and Kiya. H, “An
Encryption-then-Compression system for JPEG 2000 standard”
in IEEE Int. Conf on Acoustics, Speech and Signal Processing
(ICASSP) , pp. 1226-1230, 2015
In addition to compression ratio, PSNR is also used as a
performance metric for lossy compression of encrypted images.
Methods used in [6] , [7] , [9] , [12] and [13] uses PSNR value
as performance metric. The comparison of this is shown as
below in fig.9.
Fig. 8. Comparison graph for PSNR values
PSNR values are inversly propotional to the bit error rate.
So high PSNR values provide better result. By comparing these
methods the PSNR values ranges from 30 dB to 39 dB.
Compared to [6] and [12], methods used in [7], [9], and [14]
had high PSNR values, which implies the good quality of the
image in the reconstruction phase. This is because of the
compex processing like iteartive process for reconstruction of
images[7], multilayer decomposition[9] and wavelet
transformation[14] used in the compression of encrypted
images.
IV. CONCLUSION
This paper discussed various techniques available in ETC
system to compress encrypted image by both lossless and lossy
compression of bi-level, gray scale, binary and color images.
Using Classical compression techniques it is impossible to
compress the encrypted images. Methods discussed in [1-14]
provide solution to this by considering distributed source
coding, block based compression and prediction algorithms for
compression and simple encryption method using permutation
26
A Survey on Assistive Technology using Natural
User Interface(NUI) computing to support and
augment therapeutic rehabilitation
S.J. Syed Ali Fathima
Assistant Professor, Dept. of CSE
Kumaraguru College of Technology
Coimbatore, Tamil Nadu, India
syedalifathima.sj.cse@kct.ac.in
Dr. Shankar.S
Professor, Dept. of CSE
Sri Krishna College of Engineering
and Technology
Coimbatore, Tamil Nadu, India
shankars@skcet.ac.in
Professor,
The Oxford College of
Physiotherapy
Bangalore, Karnataka, India
ahamptsp@gmail.com
The aim of therapeutic rehabilitation is to restore patient’s
ability to independently perform basic physical activities and
their task by doing certain prescribed exercises regularly and
repeatedly. Physical therapy can help improve joint function
by focusing on range of motion exercises while decreasing
pain, swelling, and stiffness.
Abstract—Therapeutic rehabilitation is a specialty in medical
field that deals with diagnosis, evaluation, treatment and
management of people with all ages with physical or
psychosomatic disabilities due to congenital disorder, accidents
or aging problem. It deals with deduction and reduction in
disabilities by providing improvement and restoration of
movement and functional ability through regular and repetitive
physical therapy exercises continued after discharge from the
hospital. However, the efficient treatment sessions are not
guaranteed due to lack of therapists and facilities, patients were
alone for over 60% of the day, patients engaged in ‘activity’ for
only 13% of the day, undergoing some traditional therapies make
patients to lose their interest and motivating patients to continue
the exercises is lagging, which in turn makes longer time for
recovery. Thus, there is a need to find ways of cost effective,
engaging and motivated training to support and improve
recovery and rehabilitation. The focus is to use technology as a
solution involving various computing techniques as a
supplementary treatment to traditional rehabilitation and
continued assessment of disabled patients. Natural User Interface
(NUI) is the emerging technique with the ability to interact with
computers or smart devices using the human body. NUI
computing is powered by human touch, gesture, voice, thoughts
and senses. This paper is a survey on assistive technology using
emerging NUI computing techniques like touch computing,
gesture computing, surface computing, brain computing and
applications of virtual reality and augmented reality to support
and augment therapeutic rehabilitation.
The disabilities due to post-stroke, post-traumatic
paralysis, myopathy, aging etc. cannot be cured within few
days’ of hospitalization or by taking oral medicines. In some
cases, in order to make them functionally enable and improve
quality of the life, it needs continuous physiotherapy treatment
even after the discharge from hospital.
The critical issues with conventional therapy is that the
efficiency of the treatment sessions are not guaranteed due to
lack of therapists and facilities, patients were alone for over
60% of the day, patients engaged in ‘activity’ for only 13% of
the day[1] , undergoing some traditional therapies make
patients to lose their interest and motivating patients to
continue the exercises is lagging. Therefore, the recovery
takes longer period. It is been proved that 2-3 hours of arm
training a day for 6 weeks improved motor recovery when
started 1-2 months after stroke[2] .Thus, there is a need to find
ways of cost effective, engaging and motivated training to
support and improve recovery and rehabilitation.
The focus is to use technology as a solution involving
various computing techniques as a supplementary treatment to
traditional rehabilitation and continued assessment of disabled
patients. The integration of highly interactive and immersive
technologies into rehabilitation has the definite possibility to
benefit both patients and therapists.
Index Terms—Rehabilitation; CLI, GUI, NUI; Gesture
Computing; Touch Computing; Brain Computing; Surface
Computing; Virtual Reality; Augmented Reality; Processing;
Open NI
I.
Mr. A Ahamed Thajudeen
Natural User Interface (NUI) is the emerging technique
with the ability to interact with computers or smart devices
using the human body. NUI computing is powered by human
touch, gesture, voice and senses. The application of interactive
and entertaining aspects of games as exercises using NUI
facilitates rehabilitation by engaging users in a training
environment where programming can be customized or
personalized tailored to personal abilities and requirements.
This helps to motivate the patient and create a positive impact
INTRODUCTION
Rehabilitation is the process of aiding an individual to
achieve the maximum level of independence in all level such
as physically, emotionally, socially, and spiritually and
maintain the quality of life in the society. Rehabilitate is the
word from Latin “habilitas” which means “to make able
again”. Assistive technology means any system or equipment
that is commonly used to support, enhance, maintain, recover
or improve functional capabilities of person with disabilities.
27
The Comparison of computing interfaces is given in
Table1.
and enjoyable experience, which in turn to leads to improved
results and faster recovery.
The forth coming sections will review the assistive
technology using emerging NUI computing techniques like
touch computing, gesture computing, surface computing, brain
computing and NUI application technologies like virtual reality
and augmented reality to support and augment therapeutic
rehabilitation.
II.
Computing
Input data
Interfaces
Support for
Therapeutic
Rehabilitation
COMPUTING INTERFACES
CLI
Text Commands
Low
GUI
Graphical commands using
Medium
pointers and pointing
A. Command Line Interface(CLI)
CLI is also known as command language interpreter,
console user interface and character user interface (CUI) in
which user issues text commands to interact with computer
programs. CLI is mostly preferred by advanced programmers
as they are keen and strong in controlling a program or
operating systems using crisp commands.. Scripting languages
like Perl, python and ruby are CLI based programming
languages which are widely used in medical and bioinformatics to access and manage medical datasets, disease
diagnosis, drug design, evaluate treatment plan and medical
discovery[3]
devices
NUI
Speech, Body Gestures,
High
brain signals, touch,
object’s skeletal structure
Table 1: Comparison of Computing Interfaces
III.
NUI BASED COMPUTING TECHNIQUES
The major NUI based computing can be classified as
follows and the review of application of these computing
techniques in therapeutic rehabilitation is discussed.
B. Graphical User Interface (GUI)
GUI is the program interface that allows users to interact
with systems through graphical icons and visual markers or
indicators taking the advantage of computers graphics
capabilities.GUI can free the user from learning
complex command languages. GUI uses basic components
like pointers, pointing device, icons, desktop, windows and
menus. The systems using GUI are also known as WIMP
(windows Icons Menus and Pointers) systems. The invention
of various hardware devices like speech synthesis, tactile
displays, sensors etc. and appropriate software applications
enabled GUI systems to be useful in medical rehabilitation for
blind and disabled people for the improvements in orientation
and mobility. GUI has now become more significant for a
rehabilitation device. The existence of such interface improves
the user friendliness and the safety of the device as it
facilitates the control over the device and the communication
between the user and the robot[4].
a.
Touch Computing
b.
Surface Computing
c.
Gesture Computing
d.
Brain Computing
A. Touch Computing
Touch computing enables the users to interact with
controls and applications more intuitively using their touch
input. For example, touching a graphical icon to open an
application. Most popularly used touch devices are mobile
devices, tablets, touch tables and walls. The basic working
principle behind touch computing is these handheld
technology devices are built with the panel carrying an
electrical charge and when a finger touches the screen; the
panel's electrical field is disrupted by the touch input. This
disruption is recorded as a computer event and sent to the
software program for processing to initiate a response to the
touch event. Optical touch technology functions in the similar
way when a finger or an object touches the surface, it causes
the light to scatter and the reflection is caught with sensors or
cameras. These sensors send the data to software which in turn
produces the response to the touch based on the type of
reflection measured.
C. Natural User Interface (NUI)
NUI is an advancement and specialized form of GUI
which allows human-computer interaction through intuitive
actions which are common human behavior. NUI allows users
to interact with technologies in natural ways using their
gestures, senses, thoughts and speech. NUI can be operated in
many ways depending upon the purpose and user
requirements. NUI implementation can be done using
intermediary devices for interaction that are either visible or
invisible. This allows computers to be used in new areas of
applications like medical rehabilitation which has been
difficult previously. The introduction of the natural interaction
modalities has increased the attractiveness and intuitiveness of
the prototyped Serious Game for Rehabilitation area of
application [5]
Touch panel ad-hoc developed games are introduced in
cognitive exercise therapy, motivating the user to actively
participate in the rehabilitation and making the whole process
reliable and now a day’s touch screens are commonly used in
daily life, re-moving the need for an initial learning stage [6].
Multi-touch tabletop system, the AIR Touch, is developed
28
which combines existing multi-touch technologies with a suite
of new rehabilitation-centric applications with the objectives
to ensure ease of use by patients, record performance measure
and leverage therapist expertise and their knowledge of a
patient[7].. MTmini Package is useful for creating our own
touchpad using Front Diffused Illumination technique.
2D camera to identify hand gestures in 3D space is presented
as recuperation aid for younger children with impaired
movement of the upper limbs on a standard Android tablet
device [11]. A self-monitoring exercise for post-stroke sufferers
using Kinect is developed by a team at Roke Manor and
Southampton University [12].
B. Surface Computing
Surface computing is the specialized GUI that interacts
with the user through the surface of an ordinary object instead
of monitor, mouse and keyboard. Touch computing deals with
flat surface, where as surface computing concentrates on nonflat three-dimensional objects surfaces like spherical,
cylindrical and parabolic surfaces. It provides a means to
Ubiquitous Computing of making every surface as interactive
in our day to day environment. It can be implemented using
displays like LCD or Projection screens, projector and Infrared
cameras. Infrared cameras are main component that operate
independent of light and allows for gesture detection in all
lighting conditions. Microsoft research is on 3D surface
computing which deals with sensing of 3D input data above
2D surfaces. The research is being performed on the
application of surface computing for the rehabilitation of
patients with brain damage following accidents or strokes. The
computer-aided seating system (CASS) with contour sensing
and evaluation devices will facilitate the study of tissue
distortion at the loaded seat interface. This system assist in
clinical seat contour design [8].
D. Brain Computing
A brain–computer interface (BCI) is also known as mindmachine interface (MMI), direct neural interface (DNI),
or brain–machine interface (BMI), that allows direct
communication pathway between an wired brain signals and
an external device. BCI is useful in researching, assisting,
augmenting, restoring human cognitive or sensory-motor
functions. BCI read neural signals and send to computer
programs to translate input signals into action. BCI using the
thought processing offers a means for paralyzed to operate a
computer, or prosthetic limb, or motorized wheelchair.
The Comparison of NUI based computing techniques is
given in Table2.
NUI based
Input Data
Computing
Support for Therapeutic
Rehabilitation
Techniques
Touch
Touch / Multi
Simple Hand gesture
Computing
Touch on Flat
Exercises
surface
C. Gesture Computing
Gesture computing enables interface with computers using
gestures of the human body typically hands, legs, eyes, facial,
speech expressions and any body movements. Speech
computing is the part of gesture computing deals with Speech
or voice recognition that allows users to interact with a system
through spoken commands. The system recognizes the spoken
words and phrases and translates them to a machine-readable
format for interaction. In gesture recognition technology, the
movements of the human body is read by the camera and
sends the data to a computer program expecting the gestures
as input to control devices or applications. It can be 2D-based
or 3D-based, working with the help of a camera-enabled
device, which is placed in front of the individual. The cameraenabled device beams an invisible infrared light on the
individual, which is reflected back to the camera and onto
gesture recognition Integrated Chip (IC) [9]. For Example,
Camera Mouse is a program that enables to control the mouse
pointer on the computer screen by moving user's head [10].
Surface
Touch/Multi
Advanced Hand gesture
Computing
Touch on Flat or
Exercises and supports
Non-flat surface
design of surfaces for
assistive devices
Gesture
Speech, Body
Whole body gesture
Computing
Movements (or)
exercises
Gestures, Facial
expressions,
skeletal structure
Brain
Thoughts as brain
Exercises relating
Computing
signals
physical and metal
thoughts (or) controlling
Kinect is a popular motion sensing
input
devices by Microsoft for Xbox 360 and Windows PCs . The
device consists of an RGB camera, depth sensor and multiarray microphone along with a proprietary software,]which
provide
full-body
3D motion
capture, facial
recognition and voice recognition capabilities. Handpose is an
another new innovation by Microsoft Research, giving
computers the ability to accurately and completely track the
precise movement of hands including the finger wiggle
through a Microsoft Kinect. An efficient method for using a
physical gestures by
thoughts (brain signals)
Table 2: Comparison of NUI based Computing Techniques
BCI system applications can be either invasive or noninvasive. The invasive requires the direct implantation of
29
needs more assistance for training. AR technology supports
rehabilitation by providing patients with
electrodes in the user’s brain and in non-invasive, the system
captures brain signals from the electrodes attached to the
patient’s scalp through an electroencephalogram (EEG)
recording. It is harmless to use noninvasive systems to avoid
risk associated with any brain injuries. Hence, non-invasive
BCI systems are used in practical neurological rehabilitation.
Thought-based control of a neuroprosthesis is been developed
in which hand grasp with implanted stimulation electrodes [13].
Post stroke rehabilitation with feedback training using virtual
hands is been developed in which BCI use the classified brain
patterns to generate opening and closing movements of the
right or left (virtual) hand[14]. A BCI system converts the user
thoughts into actions without involving voluntary muscle
movement. Thus this system provides a new means of
communication for the patients with paralysis or severe
neuromuscular disorders [14].
IV.
an entertaining and natural environment for treatments[16] . It is
used to create more realistic and interesting exercises. It is
capable of adopting a tangible object into the framework and
enable patients to touch and interact with the real and virtual
environment all together. AR Exercises can be designed for
rehabilitating finer movements such as reach, grasp,
manipulation and release of objects. Using AR therapy with
simple assistive devices seems to be a cost effective
alternative to other forms of therapy like robotic therapy, and
also can be easily integrated with conventional physiotherapy.
V.
Processing is
an open-source programming
language and integrated development environment (IDE) to
develop programs for visual design or visual arts with images,
animations and interactions[17] . OpenNI or Open Natural
Interaction is an open source soft ware development kit (SDK)
used for the development of 3D sensing middleware libraries
and applications[18] . The OpenNI framework provides a set
of open source Application Interfaces (APIs). The APIs
provide support for speech recognition, Hand gestures and
Body Motion Tracking, OpenNI APIs can be used with
processing programming language to create NUI based games
or exercise to support therapeutic rehabilitation.
TECHNOLOGIES WITH APPLICATION OF NUI
This section reviews on two major technologies with the
application of NUI that widely supports medical rehabilitation.
a.
Virtual Reality (VR)
b.
Augmented Reality (AR)
PROGRAMMING SUPPORT FOR NUI
A. Virtual Reality(VR)
Virtual reality (VR) refers to the 3D environment generated
virtually by computers that can be explored, interacted and
experienced by the humans directly using their sight and sound
senses with the help of NUI. The environment is created using
the software and presented to the user which makes the user to
feel the real environment. The person in this environment
becomes a part of it and they are able to perform actions and
manipulate objects. The researches is being performed on
developing VR based exercises to support rehabilitation and
are proven by research to deliver significant therapeutic
benefits to patients suffering from cerebral palsy, traumatic
brain injury, stroke, autism and other physiotherapy exercise
conditions. VR can be used to develop customized exercises to
meet the target of specific therapy. Interactive Virtual Reality
Exercise System (IREX®™) is a stimulating computergenerated virtual reality therapy world developed to guide
patients that can target specific body parts through clinicianprescribed interactive rehabilitation exercises, games and
activities [15] .
CONCLUSION
In this paper we have discussed different types of
interfaces, NUI based computing techniques like touch
computing, surface computing, gesture computing and brain
computing, and the technologies like VR and AR with the
application of NUI, aiding the assistive technology to support
and enhance therapeutic rehabilitation. Numerous researches
are being carried out on these computing techniques to make it
practical in medical practice. The research should also focus
on developing suitable protocols for creating and delivery of
rehabilitation services and products with these computing
techniques to meet medical device standards.
REFERENCES
[1] Bernhardt J, Dewey H, Thrift A, Donnan G. “Inactive and alone:
physicalactivity within the first 14 days of acute stroke
unitcare.” Stroke 2004 Apr;35(4):1005–9.
[2] Lawrence ES, Coshall C, Dundas R, Stewart J, Rudd AG,
Howard R, et al. “Estimates of the prevalence of acute stroke
impairments and disability in a multiethnic population”. Stroke
2001 Jun;32(6):1279–84.
[3] Carter AB. Review of methods in medical informatics:
“Fundamentals of healthcare programming in Perl, Python and
Ruby” by Jules J. Berman. J Pathol Inform 2011;2:49
[4] Laura De Rijcke, “Development of a Graphical User Interface
for a Rehabilitation Exoskeleton”,Vrije Univeriteit Brussel,
2012-2013
[5] Rego, P.A.; Moreira, P.M.; Reis, L.P., "Natural user interfaces
in serious games for rehabilitation," in Information Systems and
Technologies (CISTI), 2011 6th Iberian Conference on , vol.,
no., pp.1-4, 15-18 June 2011
B. Augmented Reality(AR)
Augmented Reality (AR) is the technology that presents a
direct or indirect view of real world environment with virtual
objects augmented by computer generated sensory inputs like
graphics, voice, video or gestures using NUI. Unlike virtual
reality, augmented reality is very closer to the real world by
blending virtual objects to real world environment using
motion-tracking technologies, such as motion sensors or
marker recognition. The VR systems has few drawbacks in
practical and home use implementation such as use of bulky
hardware, patients unable to feel or touch virtual objects and it
30
[6] Fuyuki Matsushima, Roberto Gorriz Vilar, Keita Mitani,
Yukinobu Hoshino, “Touch Screen Rehabilitation System
Prototype Based on Cognitive Exercise Therapy,” , International
Conference, HCI International 2014, Heraklion, Crete, Greece,
June 22-27, 2014. Proceedings, Part II
[7] Michelle Annett, Fraser Anderson;Darrell Goertzen, Jonathan
Halton, Quentin Ranson;Walter F. Bischof, Pierre
Boulanger,”Using a Multi-touch Tabletop for Upper Extremity
Motor Rehabilitation”, OZCHI 2009 Proceedings ISBN: 978-160558-854-4
[8] D. M. Brienza, K.-C. Chung, C. E. Brubaker, and R. J.
Kwiatkowski,”Design of a Computer-Controlled Seating
Surface for Research Applications”, IEEE transactions on
Rehabilitation Engineering, vol. 1, no. 1, march 1993
[9] http://www.technavio.com/blog/top-18-gesture-recognitiontechnology-companies
[10] http://www.cameramouse.org/
[11] E. Ziogas, J. Eyre,G. Morgan, “An Efficient Application of
Gesture Recognition from a 2D Camera for Rehabilitation of
Patients with Impaired Dexterity”, Newcastle University,
TECHNICAL REPORT SERIES,No. CS-TR-1368 January,
2013
[12] Roke Manor Research Ltd, Microsoft Kinect® Gesture
Recognition Software For Stroke Patients,
http://www.roke.co.uk/resources/articles/2012-TTP-Kinect.pdf
[13] Aymeric Guillot, Christian Collet, “The Neurophysiological
Foundations of Mental and Motor Imagery”, Oxford University
press,2010
[14] Pfurtscheller, G.; Muller-Putz, G.R.; Scherer, R.; Neuper, C.,
"Rehabilitation with Brain-Computer Interface Systems," in
Computer , vol.41, no.10, pp.58-65, Oct. 2008
doi: 10.1109/MC.2008.432
[15] http://www.gesturetekhealth.com/products-rehab.php
[16] A. Merians, D. Jack, R. Boian, M. Tremaine, G. Burdea, S.
Adamovich, M. Recce, and H. Poizner, “Virtual realityaugmented
rehabilitation for patients following stroke,” Phys. Ther., vol. 82,
no. 9, pp. 898–915, Sep. 2002. [Online]. Available:
http://www.ptjournal. org/cgi/content/abstract/82/9/898
[17] https://processing.org/
[18] http://openni.ru/
31
Experimental Investigation on Weld
Strength of Copper Wire – Al8011 Sheet
Using Ultrasonic Metal Welding
R. Karthik
Dr. P. Ashoka Varthanan
Assistant Professor
Professor & Head
UG Student
Sri Krishna College of Engg & Tech
Coimbatore, Tamil nadu, India
karthikr@skcet.ac.in
Sri Krishna College of Engg & Tech
Coimbatore, Tamil nadu, India
ashokavarthanan@skcet.ac.in
Sri Krishna College of Engg & Tech
Coimbatore, Tamil nadu, India
12ba052@skcet.ac.in
Abstract—High frequency vibrations are produced to
weld the specimen by means of ultrasonic welding
with less amount of heat energy. The sheet metal used
in this trial analysis has normal and shear force which
acts on the parts to be welded and the weld interface.
The forces are the result of ultrasonic vibrations of
the tool, pressed onto the parts to be welded. This
paper made an attempt to the produce the good
quality weld strength using ultrasonic welding with a
comparison of experimental values and RSM
predicted values. Experiments are performed using
Al-8011 plate and copper wire. Ultrasonic welding is
completed with the response surface methodology to
study the effect of input parameter such as weld
pressure, weld time and amplitude. The welded
samples are subjected to tensile test to find out the
maximum weld strength. Finite element analysis is
carried out to analyse the stress distribution and
structural analysis.
V. S. Gowthaman
automotive industry. It is a quasi-solid-state
process that produces a weld by introducing high
frequency vibration to the weldment as it is held
under moderately high clamping forces.
Fig.1 Ultrasonic metal welding
Index Terms— ultrasonic welding, Al8011(Aluminium
alloy) weld strength, response surface methodology
(RSM).
II.LITERATURE SURVEY
S. Elangovan, et al [1] described the joining of
two materials together using high frequency
vibrations in ultrasonic welding without producing
significant amount of heat. During ultrasonic
welding of sheet metal, normal and shear forces act
on the parts to be welded and the weld interface.
These forces are the result of ultrasonic vibrations
of the tool, pressed onto the parts to be welded.
They carried out a study on model for the
temperature distribution during welding and stress
distribution in the horn and welded joints are
presented. The presented finite element model is
capable of predicting the interface temperature and
stress distribution during welding and their
influences in the work piece, sonotrode and anvil.
Ganesh M, et al [2] stated that Ultrasonic
welding is devoted to weld thin sheet metals of
similar or dissimilar couples of non-ferrous alloys
I. INTRODUCTION
A. Ultrasonic Welding
Ultrasonic Welding process, in which two work
pieces are bonded as a result of a pressure exerted
to the welded parts combined with application of
high frequency acoustic vibration (ultrasonic).It
causes friction between the parts, which results in a
closer contact between the two surfaces with
simultaneous local heating of the contact area.
Interatomic bonds, formed under these conditions,
provide strong joint. Thickness of the welded parts
is limited by these power of the ultrasonic
generator. It is used mainly for bonding small work
pieces in electronics, for
manufacturing
communication devices, medical tools, watches, in
32
like copper, aluminium and magnesium without
addition of filler material resulting in high quality
weld; it can count on a low energy consumption
and on a joining mechanism based on a solid state
plastic deformation which creates a very
homogeneous metallic structure between the base
materials, free from pores & characterized by
refined grains and confined inclusions. It can join
also painted or covered sheet metals. Thin sheets of
aluminium have been joined to thin copper sheets
by means of Ultrasonic spot Welding. Results are
particularly effective in order to evaluate the
relevance of various phenomena influencing the lap
joint technique obtained on thin aluminium and
copper by the application of Ultrasonic Metal Spot
Welding (USMSW).in this we also observed that
when ultrasonic welding done on a specimen in
three different parameters, the weld time and
amplitude is playing vital role in weld strength.
When amplitude is low with high weld time and
moderate pressure, weld strength is poor. When
pressure is low high amplitude and high weld time,
the weld strength is good. When the amplitude is
high, the weld strength also is high.
S. Elangovan, et al [3] focuses on the
development of an effective methodology to
determine the optimum welding conditions that
maximize the strength of joints produced by
ultrasonic welding using response surface
methodology. RSM is utilized to create an efficient
analytical model for welding strength in terms of
welding parameters namely pressure, weld time,
and amplitude. Experiments were conducted as per
central composite design of experiments for spot
and seam welding of 0.3- and 0.4-mm-thick Al
specimens. An effective second-order response
surface model is developed utilizing experimental
measurements. Further, it is concluded that weld
strength decreases with increase of pressure
because increase in clamping force (pressure)
reduces the relative motion between surfaces
leading to reduced area of contact and hence
reduced strength. Also, weld strength increases
with increase of amplitude because increase in
amplitude gives increased area for rubbing action
between the metallic surfaces that leads to better
bonding and increase of weld strength. In some
cases, weld strength increases considerably up to
2.5 s. Beyond 2.5 s, the weld strength again starts
decreasing for any value of pressure and amplitude
because excessive weld time affects the existing
molecular bond in the weld joint.
V. Gunaraj, et al[4] describes Response
surface methodology (RSM) is a technique to
determine and represent the cause and effect
relationship between true mean responses and input
control variables influencing the responses as a two
or three dimensional hyper surface. selection of the
optimum combination of input variables for
achieving the required qualities of weld. This
problem can be solved by the development of
mathematical models through effective and
strategic planning and the execution of experiments
by RSM. This paper highlights the use of RSM by
designing a four-factor five-level central composite
rotatable design matrix with full replication for
planning, conduction, execution and development
of mathematical models. These are useful not only
for predicting the weld bead quality but also for
selecting optimum process parameters for
achieving the desired quality and process
optimization. RSM can be used effectively in
analysing the cause and the effect of process
parameters on response. The RSM is also used to
draw contour graphs for various responses to show
the interaction effects of different process
parameters.
Marius Pop-Calimanu, et al [5] stated that
the Ultrasonic welding is a solid-state welding
process in which similar or dissimilar work pieces
are jointed by the application of high frequency
vibratory energy, where the work pieces are held
together under pressure without melting. The
problem faced by researchers and industry which
have deals with ultrasonic welding process is poor
strength of the weld, due to improper selection of
welding parameters. Therefore, we are usually
interested to determine which variables process
affects the response. A stage is to optimize, this
means that we can determine the region in the
important factors that lead to the best possible
response. In this paper, welding parameters, like
welding time, welding pressure and amplitude of
the vibration are taken into account during the
realization of ultrasonic welded joints of
Al/20%SiC composite material under disks form,
whose thickness are 1 mm. This work focuses on
the development of an effective methodology to
determine the optimum conditions for welding, that
maximize the strength of joints produced by
ultrasonic welding. It was observed that welding
pressure and welding time has a significant effect
on the response (joint strength). It was observed
that the interactions of influence factors are more
significant on the net area of the joint, than on the
energy input on joint formation.
J. Pradeep Kumar, et al [6] states that
ultrasonic welding is as solid state joining process
that produces joints by the application of high
frequency vibratory energy in the work pieces held
together under pressure without melting. In
electronic and automotive applications, copper
wires are connected to the equipment (alternator/
rectifier) by a solid state joining process. For such
an application ultrasonic metal welding is useful.
This paper presents a study of ultrasonic welding of
copper wire to thin copper sheet as many of the
industrial applications are in need of this kind of
contact technology. Quality of the welded joints is
evaluated based on mechanical tests and the quality
33
criterion is then applied to evaluate the weld
strength. A second order regression model equation
is developed to predict the weld strength of the
joint based on the experiments conducted using full
factorial design of experiments. Experimental study
conducts for ultrasonic metal spot welding and
defines the quality of the joint using maximum
weld strength and failure type in the tensile test.
For this study, copper sheet and copper wire were
selected because they are the materials and joints
effectively used for many of the industrial
applications. The experiment was conducted using
full factorial design according to varying weld
pressure, amplitude and welding time. The standard
tensile test was selected to extract quality response.
Using ANOVA the percentage contribution of each
factors are determined. It can be found that the
second order regression model equation derived
from present study can be readily applied to predict
the weld strength value of the joint made between a
copper sheet and a copper wire.
potentially impact one or more Response variables.
RS model, which is an analytical function in
predicting weld strength values, is developed using
RSM. The second-order mathematical models have
been developed to predict the weld strength. The
polynomial equation for the three factors
considered in the present case is
Where
represents pressure, weld time, and amplitude; ,
,
, and
represent the constant, linear,
quadratic, and interaction terms, respectively. The
weld strength obtained from experimental results
for different combinations of parameters is given as
input to the Minitab software, and a second-order
mathematical model for predicting weld strength is
developed.
III. MATERIAL PROPERTIES
The material used here for ultra sonic welding is Al
8011, due to its properties of wrought, low weight,
corrosion resistance, and easy maintenance of final
product, The material and its alloys has a long
duration. The other material used in ultrasonic
welding is copper. The chemical composition listed
in the Table I.
V. METHODOLOGY
A. Methodology for experimental work
The material and the range of weld
parameters selected.
The weld specimens are prepared
according to ASTM standards (D100201).
Table I. chemical composition of al 8011
Element
Al
Fe
Si
Mn
Zn
Cu
Ti
Cr
Mg
is the response, i.e., weld strength;
Content(%)
97.3-98.9
0.60-1
0.50-0.90
0.20
0.10
0.10
0.080
0.050
0.050
The experiments are conducted as per
Design of experiment (DOE) using
Response surface method (RSM).
The weld strengths for each case are
observed and compared with analyzed
results.
B. Methodology for analysis work
A 3D model of the weld specimen is
created.
IV. RESPONSE SURFACE METHODOLOGY
The element type and material properties
for stress analysis are selected.
Response surface methodology (RSM) is defined as
a collection of mathematical and statistical methods
that are used to develop, improve, or optimize a
product or process. The method was introduced by
Box and Wilson. The main idea of RSM is to use a
sequence of designed experiments to obtain an
optimal response. It comprises statistical
experimental designs, regression modelling
techniques and optimization methods. Most
application of RSM involves experimental
situations where several independent variables
The
initial
conditions,
conditions are selected.
boundary
Various loads are given at the appropriate
nodes of the model and analysis is carried
out.
The simulation results are compared with
the experimental results obtained.
34
Make: Metal weld 2500, National Indosonic,
Pune
VI. EXPERIMENTAL SETUP
The experiment analysis was done with a
conventional ultra sonic welding machine
involving different weld parameters. The ultrasonic
welding machine used has got a horn made of
hardened steel (spot type) and anvil made of steel
with serrations at the top surface. The horn area
which comes in contact has serrations similar to the
top surface of the anvil to prevent the work piece
from sliding during welding. The specimen (0.3
mm Al 8011 sheet and 1.2mm diameter copper
wire) were specimens. Welding experiments are
conducted by considering three welding parameters
namely pressure, weld time, and amplitude of horn
vibrations. A total of 20 welds with 20 different
combinations of different pressure, weld time and
amplitude are carried out. Low–middle–high level
of welding parameters in welding space for
response surface methodology design is shown in
Table II. Ranges of welding parameters are selected
based on trial and error method.
Table III. Ultrasonic Welding Machine Specification
Input power
Output power
Output frequency
Maximum pressure
Stroke length
Maximum Amplitude
230V, 50Hz single phase
2500 W
20 KHz
8 bar
25mm
60 μm
B. Tensile testing
The welded samples are subjected to tensile test in
universal testing machine. 20 welded samples are
tested then the maximum load that the weld
specimen can withstand are obtained.
Table II. Experimental parameters and their levels
S. No
Factors
Designation
Level
1
Level
2
Level
3
1
Pressure
(bar)
A
2.5
3
3.5
2
Time (sec)
B
2
2.5
3
3
Amplitude
(µm)
C
28
42.5
57
Fig. 3 universal testing mahine
Table IV. Specifications of universal testing machine
Model
Maximum capacity
Least count
displacement
Grip separation
TKG – EC – 10KN
10KN
0.1mm
25 – 750mm
Fig.2 Welded specimens
A.ULTRSONIC
SPECIFICATION:
WELDING
MACHINE
Fig.4 Welded specimens after testing
35
Table V. ANOVA result on effect of process parameter on
VII. RESULTS AND DISCUSSIONS
weld strength
A. Effect of process parameters on weld strength
The process parameters such as weld pressure,
weld time and amplitude has significant effect on
weld strength was shown in Figure 5. Table V
represents the ANOVA results of weld strength. It
could be found from the figure and table that
amplitude of horn vibration has greatest effect on
weld strength by contribution of 19.13%. Also the
P-value reports that this factor is really very
significant.
It could be found from the figure 5 and table V that
weld pressure has effect on weld strength which
increases the weld strength in accordance with
pressure and a proficient gain of 13.56% of weld
strength. Also table proves that weld pressure has
significant effect on weld strength and the weld
time has not significant as like amplitude of the
horn vibration and the weld strength.
Main Effects Plot for WS
Fitted Means
P
T
DF
9
Adj SS
21.6995
Adj MS
2.41105
F Value
Model
Linear
3
9.1117
3.03725
10.91
P
1
3.7758
3.77576
13.56
0.01071
0.04
8.66
T
1
0.01071
A
1
5.3253
5.32527
19.13
Square
3
12.2692
4.08973
14.69
P*P
1
1.0825
1.08248
3.89
T*T
1
0.0731
0.07314
0.26
A*A
1
2.6797
2.67972
9.63
2-Way
3
0.3186
0.10619
0.38
P*T
1
0.2106
0.21064
0.76
P*A
1
0.0801
0.08014
0.29
0.02779
0.10
T*A
1
0.0278
Error
10
2.7840
0.27840
Lack-of-Fit
5
1.9057
0.38115
2.17
Pure Error
5
0.8782
0.17565
2.17
Total
19
24.4835
A
21.5
The response surface is plotted to study the effect
of process variables on the weld strength and is
shown in Figures 6-8. From Figure 6 the weld
strength is found to be increasing steeply in
accordance with the increase of weld pressure and
increase of weld time respectively. It is observed
from the Figure 7 that the weld strength increases
with increase in weld pressure and amplitude of
horn vibration. The Figure 8 shows that the weld
strength increases with increase of amplitude of
horn vibration and slightly increases in weld time.
21.0
M ean of WS
Source
20.5
20.0
19.5
2.5
3.0
3.5 2.0
2.5
3.0 30
40
50
Fig.5 effect of process parameters on weld strength
Surface Plot of WS vs T, P
Hold Values
A 42.5
2 1 .0
WS 20.5
20.0
3.0
1 9 .5
2.5
2.5
2
.5
P
3 .0
3 .5
T
2.0
Fig.6 Combined effect of pressure and time on weld strength
36
Surface Plot of WS vs A, P
8
2.5
3.0
28.0
19.9299
19.83427
9
3.0
2.5
57.0
21.5909
21.28843
10
3.0
2.5
28.0
20.2513
19.82895
11
3.5
3.0
28.0
21.1400
21.58791
12
3.0
3.0
42.5
20.4120
19.76736
13
3.0
2.5
42.5
19.0069
19.57155
14
3.0
2.5
42.5
18.8123
19.57155
15
3.0
2.5
42.5
19.5883
19.57155
16
3.5
2.0
57.0
22.1804
22.45726
17
3.5
2.0
28.0
21.2163
21.08007
18
2.5
2.0
57.0
22.0197
21.75302
19
3.0
2.5
42.5
19.9767
19.57155
20
3.0
2.0
42.5
19.7821
19.70192
Hold Values
T 2.5
22
WS
21
60
20
50
19
A
40
2.5
2
.5
P
3 .0
30
3 .5
Fig.7 Combined effect of pressure and amplitude on weld
strength
Surface Plot of WS vs A, T
Hold Values
P 3
2 1 .6
WS
20.8
60
20.0
50
40
2.0
2
.0
T
2 .5
A
30
3 .0
Fig.8 Combined effect of time and amplitude on weld strength
A total of 20 experiments were conducted at
different levels of parameters to obtain ultrasonic
spot welded joints of 0.3 mm Aluminium 8011
sheet and 1.2mm diameter copper wire. The values
of weld strength obtained from experiments and
those predicted from response surface model along
with design matrix are tabulated in Table VI for
spot welding of 0.3mm-thick Al 8011 and 1.2 mm
dia copper wire specimens.
Figure 9 illustrates the comparison between the
experiment sets and RSM predication for weld
strength and it’s found that the experimental and
predicted values are equal this shows the reliability
of the RSM models.
While comparing the
predicted values of strengths with experimental
values, it was observed that the deviations are
minimum except for a few combination of
parameters.
Table.VI List of RSM prediction with experimental dataset
Avg. Strength (N/mm2)
S.
Pressure
Time
Amplitude
No
(bar)
(sec)
(µm)
1
3.0
2.5
42.5
19.3945
19.57155
2
2.5
3.0
57.0
21.0587
21.37605
3
2.5
2.0
28.0
19.7692
19.97548
4
3.5
2.5
42.5
21.3769
20.81342
5
3.5
3.0
57.0
22.7544
22.72935
6
2.5
2.5
42.5
19.7458
19.58448
7
3.0
2.5
42.5
19.2007
19.57155
Experimental Predicted
Fig.9 experiment sets vs. RSM prediction for weld
strength
37
B. Finite element analysis for stress distribution
The finite element analysis was used to
determine the stress distribution for al 8011 to
copper welding. The contact between the two work
pieces is established and transient analysis is done
in order to find out the stress distribution in the
weld. Figure 10 shows the finite element model
used in the welding of the Al8011-copper
specimen. It has quadratic displacement behaviour.
The element is defined by ten nodes having three
degrees of freedom at all nodes. The element has
plasticity, creep, stress stiffening, large deflection,
and large strain capabilities. The pressure is applied
on the top surface of the top part on the nodes of
the area only where the horn comes into contact
with the work piece and the resulting displacement
of the work piece on top due to the high-frequency
vibration of the horn are fed as inputs for the
structural analysis. The work piece at the bottom is
arrested in all degrees of freedom.
Fig.11 Stress distribution on spot welded work piece
VIII.CONCLUSION AND FUTURE WORKS
The process of ultrasonic metal welding
and the mechanism of bond formation
during welding are studied. Ultrasonic
welding of copper wire to al 8011 sheet
joints are performed successfully and the
parameters affecting ultrasonic metal
welding are studied.
The material properties considered were Young’s
modulus (E), Poisson’s ratio (μ) and density (ρ) for
performing structural analyses as shown in Table
VII. The stress distribution was obtained according
to the tensile load applied and that finite element
model after stress distribution was shown in the
Figure 10.
Based on the design matrix developed
using response surface methodology
(RSM) the optimum combination of weld
parameters for copper wire to al 8011
sheet joint was identified.
The difference between weld strengths
obtained from experiments and those
predicted by RSM is minimum was
observed.
Increasing usage of light weight composite
materials in aerospace and automotive
industries are preferring ultrasonic
welding.
Fig.10 finite element model for spot welded work piece
IX. REFERENCES
Table VII. Material properties of work piece
Material
Young’s
modulus
(GPa)
Poisson’s
ratio
Density
(Kg/m3)
Al 8011
70
0.35
2710
Copper
120
0.34
9100
[1] S. Elangovan, S. Semeer, K. Prakasan, "Temperature
and stress distribution in ultrasonic metal welding—An
FEA-based study", journal of materials processing
technology 209 (2009) 1143–1150
[2] Ganesh M, Praba Rajathi R, "Experimental study on
ultrasonic welding of Aluminium sheet to copper sheet",
International Journal of Research in Engineering and
Technology, 02 (2013) 161-166
[3] S. Elangovan, K. Anand, K. Prakasan, "Parametric
optimization of ultrasonic metal welding using response
surface
methodology and
genetic
algorithm",
38
International Journal of Advanced Manufacturing and
Technology 0.1007 s00170-012-3920
[4] V. Gunaraj , N. Murugan, "Application of response
surface methodology for predicting weld bead quality in
submerged arc welding of pipes", Journal of Materials
Processing Technology 88 (1999) 266–275
[5]Marius Pop-Calimanu, Traian Fleser, "The Increasing
of weld strength by parameters optimization of
Ultrasonic welding for composite material based on
Aluminium using design of experiments", nanocon, 2012
[6]J.Pradeep Kumar, K.Prakasan, "Experimental studies
on joining Copper wire - Copper sheet using Ultrasonic
Metal welding", International Journal of Mechanical and
Production Engineering Research and Development
(IJMPERD ) ISSN 2249-6890 Vol.2, Issue 3 Sep 2012
21-29
39
Effect of Vibratory treatment on Metallurgical and
Mechanical properties of Aluminum alloys
weldments
Dr.K.Balasubramanian
Dr.V.Balusamy
Department of Mechanical Engineering
Sri Krishna College of Engineering and Technology
Coimbatore-641 008, Tamil Nadu, India
skcetbalu@gmail.com
Department of Metallurgical Engineering
PSG College of Technology, Coimbatore-641 004
Tamil Nadu, India
v_b_samy@yahoo.co.in
2.5%Mg and proved that the hot cracking resistant got
improved. They used Houldcroft test [8] to evaluate the hot
cracking susceptibility of the welds. Yoshiki et al [9] gained
refined crystal grains of magnesium by means of
electromagnetic vibrations during solidification of pure
magnesium. Equiaxed grains were formed when the
electromagnetic vibrations had frequencies less than 1000Hz.
Cui et al [10] used ultrasonic vibration to refine the 316L weld
metal micro structure and found columnar dendritic micro
structure decreased from 95% to 10%.
Benefits of vibration on grain refinement and mechanical
properties such as Hardness, Yield strength, Ultimate Tensile
Strength and Breaking Strength have been discussed by
different authors [11, 12 and 13]. Lu Qinghua et al [14]
applied vibration during submerged arc multi pass welding for
improving the quality of full welded valve and found
significant reductions in the welding deformation and residual
stress. Pučko [15] gained positive effect on impact strength by
means of vibration during welding. The reason is vibration
stabilizes the micro structure to become more resistant to heat
affects that could minimize impact toughness. It was also
found that the type of fracture turns more ductile with
vibration during welding.
The present work aims the improvement of metallurgical
and mechanical properties of aluminum alloy weldment by
applying vibratory treatment during the arc welding.
Abstract— In this work, an attempt is made to improve the
metallurgical and mechanical properties of high strength of
aluminum alloys through vibratory treatment. The materials
used for the investigation are AA6061 alloy and AA2024 alloy.
An important metallurgical difficulty is the formation of hot
cracking during arc welding. Experiments were carried out in
the presence and absence of vibratory treatment. Weldments
made with and without vibratory treatment were compared
using hot cracking test and characterisation tests like X-ray
diffraction
techniques
and
hardness
measurements.
Experimental results show that the metallurgical and mechanical
properties are improved by vibratory treatment.
Index Terms— AA6061 alloy, AA2024 alloy, Hot cracking, Xray diffraction, hardness
I. INTRODUCTION
An important metallurgical difficulty in arc welding of high
strength aluminum alloy is the formation of hot cracking. Hot
cracking is a high temperature cracking mechanism and is a
function of solidification behaviour of alloy system. The
formation of fine grained structure in the weld metal is one of
the important methods to control hot cracking. Many methods
are used to control the grain structure in the weld metals
which have been reported in literature and discussed below.
Yunjia et al [1] and Dvornak et al [2] studied the weld metal
grain refined by adding Ti and Zr as inoculants. Mousavi et al
[3] achieved grain refinement in electromagnetically stirred
AA7020 welds and proved that the grain refinement was due
to the grain detachment. Kou and Le [4] achieved grain
refinement in an oscillated arc weld of commercial 6061
aluminum alloy. Sundaresan and Janakiram [5] achieved
considerable refinement of the fusion zone grain structure in
α-β titanium alloys welding using this technique.
Balasubramanian et al [6] showed finer and more equiaxed
grain structure in GTA and GMA Welds of High strength
aluminum alloys by pulsed current.
II. EXPERIMENTAL PROCEDURE
A. Materials
The materials used for this investigation were AA2024
alloy and AA6061 alloy and their chemical compositions are
given in table 1.
TABLE I CHEMICAL COMPOSITION, WT.% OF ALUMINUM
ALLOYS INVESTIGATED
Alloy 1
AA2024
Some of the works related to vibratory treatment have been
reported and discussed follows. Using torch vibration, Davies
and Garland [7] produced grain refined weldment during
autogeneous GTA welding of aluminum alloy containing
Alloy 2
AA6061
40
Mg
0.83
Cu
4.36
Mg
1.06
Si
0.093
Mn
0.53
Fe
0.15
Al
Bal
Zn
0.14
Mn
0.096
Fe
0.41
Si
0.62
Cu
0.21
Cr
0.18
Al
Bal
B. Test specimen
To evaluate the specimens for resistance against hot
cracking, Houldcroft hot cracking test was employed. In 1955,
Houldcroft [8] first developed the so called fish bone test to
evaluate the hot cracking susceptibility of welds. The
specimen used for this work is of size 76 mm x 44 mm, and its
thickness is 4 mm. The specimen contains groves in two rows
with nine grooves in each row, cut to different depths, as
shown in figure 1.
vibratory treatment. The Vickers hardness test was also
conducted to assess mechanical properties.
III. RESULTS AND DISCUSSIONS
A. Effect of vibratory treatment on hot cracking
Houldcroft tests were carried out to find the resistance of
weldment against hot cracking. Specimen welded with
vibratory treatment was compared with the specimen welded
without vibration. Vibratory treatment was carried out at the
frequency 700 Hz for AA2024 alloy and 600 Hz in case of
AA6061 alloy.
Photographs showing cracks of different lengths in the
Houldcroft specimens are shown in figure 3. The hot crack
length and its sensitivity factor are shown in table 2.
Slot 0.8 mm
wide
44
76
All the dimensions are in mm
Fig.1. Houldcroft cracking test specimen
C. Welding
To carry out the hot cracking test, Gas Tungsten Arc
Welding was used. The main parameters used for welding are
Welding Current
Travel speed
Electrode
Shielding gas
Polarity
-
(a)
(b)
Fig.3 [a] Houldcroft test results for AA2024 alloy (a) without vibration
(b) with vibratory treatment of frequency 700 Hz
150 A
6.25 mm/s
EWTh-2, 3 mm diameter
pure argon gas
DCEN
D. Vibratory treatment
The vibration generator cum analyser used in this work
consists of Piezo electric transducer capable of producing
mechanical vibration in the frequency range of 100 Hz to 3000
Hz. The transducer is made to transmit the vibratory energy to
the weld plate through a welding fixture. The welding fixture
and the specimen holding method are such that the energy loss
during transmission from the transducer to the weld plate is
kept to the lower value. The arrangement of this vibratory
treatment unit is shown in figure 2.
(a)
(b)
Fig.3 [b] Houldcroft test results for AA6061 alloy (a) without
vibration (b) with vibratory treatment of frequency 600 Hz
TABLE II. CRACK SENSITIVITY OF ALUMINUM ALLOYS WITH AND
WITHOUT VIBRATORY TREATMENT
Crack length [mm]
Sl.
No
1
2
Aluminum
alloys
AA2024
alloy
AA6061
alloy
without
vibratory
treatment
50.5
37.5
with
vibratory
treatment
34.5
[700 Hz]
32
[600 Hz]
Crack sensitivity [%]
without
vibratory
treatment
with
vibratory
treatment
67.3
46
50
42.7
In AA2024 alloy, the crack sensitivity was around 67.3%
when the specimen was welded without vibration. The crack
sensitivity decreased to 46% when the specimen was welded
with vibration of frequency 700 Hz. In AA6061 alloy, the
crack sensitivity was around 50% when the specimen was
welded without vibration. The crack sensitivity decreased to
42.7% when the specimen was welded with vibration of
frequency 600 Hz.
B. Effect of vibratory treatment on grain size
X-Ray Diffraction technique was used to measure the
grain size in order to quantify the level of grain refinement
due to vibratory treatment. The grain size was measured for
the weld metals AA2024 alloy and AA6061 alloy. The
average grain size was calculated by using Scherrer’s
equation.
Fig.2. Experimental set up
E. Characterisation tests
The extent of grain refinement due to vibratory treatment
is determined through X-Ray Diffraction techniques. X-Ray
Diffraction technique was applied to the welded specimen,
mainly to compare the grain size of welded specimens with
41

where,
K
 cos
TABLE III. FULL WIDTH HALF INTENSITY AND DIFFRACTED
ANGLE FOR AA2024 ALLOY
τ – Grain size, in micron
λ – Wavelength of cobalt (1.79x10-4 micron)
β – Full width of half intensity (radian)
θ – Diffracted angle
Sl.No
1
2
3
4
5
6
Figures 4 and 5 show the X-ray diffraction patterns for the
AA2024 alloy and AA6061 alloy respectively which exhibit
significant line broadening. The extent of broadening is the
full width at half maximum intensity of the peak. From the
results of XRD technique, the values of full width of half
intensity and diffracted angle for the strongest peaks with
different frequency are tabulated in Table 3 and 4 for AA2024
alloy and AA6061 alloy respectively.
0
700
Peak
No
2θ (Deg)
FWHD
(B) (Deg)
1
5
2
3
10
7
44.96
77.3
52.4
44.93
77.26
52.34
0.155
0.167
0.176
0.238
0.291
0.250
TABLE IV. FULL WIDTH HALF INTENSITY AND DIFFRACTED
ANGLE FOR AA6061 ALLOY
Sl.No
1
2
3
4
5
6
(a)
Frequency
of vibration
(Hz)
Frequency
of vibration
(Hz)
0
600
Peak
No
2θ
(Deg)
FWHD
(B)
(Deg)
10
16
13
3
2
6
44.47
76.78
52.31
52.39
44.92
77.23
0.078
0.104
0.150
0.125
0.195
0.204
Calculations [AA2024 alloy]
Case (i) without vibration
d = (0.9 x 1.79 x 0.0001) / (0.155 x cos22.48 x 0.0175) =
0.063 micron
d = (0.9 x 1.79 x 0.0001) / (0.167 x cos38.65 x 0.0175) =
0.067 micron
d = (0.9 x 1.79 x 0.0001) / (0.176 x cos26.20 x 0.0175) =
0.057 micron
The average grain size is 0.062 micron.
(b)
Case (ii) with vibration (700Hz)
d = (0.9 x 1.79 x 0.0001) / (0.238 x cos22.26 x 0.0175) =
0.041 micron
d = (0.9 x 1.79 x 0.0001) / (0.291 x cos38.63 x 0.0175) =
0.038 micron
d = (0.9 x 1.79 x 0.0001) / (0.250 x cos26.17 x 0.0175) =
0.040 micron
The average grain size is 0.039 micron.
Figure 4. Diffraction patterns for AA2024 alloy (a) without vibratory
treatment (b) with vibrated at 700 Hz showing
broadening because of particle size
Calculations [AA6061 alloy]
Case (i) without vibration
d = (0.9 x 1.79 x 0.0001) / (0.078 x cos22.24 x 0.0175) =
0.126 micron
d = (0.9 x 1.79 x 0.0001) / (0.104 x cos38.39 x 0.0175) =
0.107 micron
d = (0.9 x 1.79 x 0.0001) / (0.150 x cos26.16 x 0.0175) =
0.067 micron
The average grain size is 0.1 microns.
(a)
(b)
Case (ii) with vibration (600Hz)
d = (0.9 x 1.79 x 0.0001) / (0.125 x cos26.20 x 0.0175) =
0.0804 micron
d = (0.9 x 1.79 x 0.0001) / (0.195 x cos22.46 x 0.0175) =
0.0503 micron
d = (0.9 x 1.79 x 0.0001) / (0.204 x cos38.62 x 0.0175) =
0.0549 micron
The average grain size is 0.062 micron.
Figure 5. Diffraction patterns for AA6061 alloy (a) without vibratory
treatment (b) with vibrated at 600 Hz showing
broadening because of particle size
42
REFERENCES
In AA2024 alloy, the average grain size was found to be
0.062 micron for the non vibrated specimen and 0.039 micron
for the specimen vibrated with a frequency of 700 Hz.
Similarly, In AA6061alloy, the average grain size was found
to be 0.1 micron for the non vibrated specimen and 0.062
micron for the specimen vibrated with a frequency of 600 Hz.
The grains of the vibrated specimen are found to be finer.
[1]
[2]
[3]
C. Effect of vibratory treatment on hardness
From the results of aforesaid tests of grain size
measurement by XRD technique, it is found that the vibratory
treatment has helped in the formation of fine grains. Fine grain
structure is known to improve the mechanical properties
particular the hardness. Weld metal hardness values were
measured for AA2024 alloy and AA6061 alloy with and
without vibratory treatment. Hardness was measured using a
Vickers Hardness tester at a load of 250 grams. The Vickers
hardness values of the weld metals AA2024 alloy and
AA6061 alloy without vibratory treatment and with vibratory
treatment are given in Table 5.
[4]
[5]
[6]
TABLE V. HARDNESS VALUES OF ALUMINUM ALLOYS WITH AND
WITHOUT VIBRATORY TREATMENT
[7]
Vickers Hardness numbers
Sl.No
1
2
Aluminum
alloys
AA2024
alloy
AA6061
alloy
without
vibratory
treatment
109
86
[8]
with
vibratory
treatment
[9]
131
[700 Hz]
93
[600 Hz]
[10]
The results show that the hardness has increased from 86
VHN to 93 VHN due to vibratory treatment for AA2024 alloy
and from 109 VHN to 131 VHN in case of AA6061 alloy.
[11]
IV. CONCLUSION
The effect of vibratory treatment on metallurgical and
mechanical properties was investigated and found that:
 The vibratory treatment applied during Gas Tungsten
Arc Welding of AA2024 alloy and AA6061 alloy
resulted in reduction of hot cracking in the weld
metal.
 Grains were disturbed and grain refinement of
AA2024 alloy and AA6061 alloy was achieved by
imposition of vibratory treatment during Gas
Tungsten Arc Welding.

The Vickers hardness test has confirmed that the
hardness values for the vibrated samples have
improved.
[12]
[13]
[14]
[15]
43
H. Yunjia, R.H. Frost, D.L. Olson and G.R. Edwards,
“Grain Refinement of Aluminum Weld Metal”, Welding
Journal, pp.280s, 1989.
M.J. Dvornak, R.H. Frost and D.L. Olson, “The
Weldability and Grain Refinement of Al-2.2Li-2.7Cu”,
Welding Journal, pp.327s, 1989
M.G. Mousavi, M.J.M. Hermans, I.M. Richardson and G.
Den Ouden, “Grain refinement due to grain detachment in
electromagnetically stirred AA7020 welds”, Science and
Technology of Welding and Joining, Vol. 8, No. 4, pp.
309-312, 2003.
S. Kou and Y. Le, “Nucleation Mechanisms and Grain
Refining of Weld Metal”, Welding Journal, pp.305s,
December. 1986.
S. Sundaresan and G.D. Janakiram, “Use of Magnetic arc
oscillation for grain refinement of gas tungsten arc welds in
α-β titanium alloys”, Science and Technology of Welding
and Joining, vol. 4, pp.151, 1999
V. Balasubramanian, V. Ravisankar, and G. Madhusudhan
Reddy,“Effect of pulsed current and post weld aging
treatment on tensile properties of argon arc welded high
strength aluminum alloy”,
Materials Science and
Engineering A, vol. 459, pp.19-34, 2007
G.J. Davies and J.G. Garland, International Metals
Review, Vol. 20(196), pp.83, 1975.
P.T. Houldcroft, “A simple cracking test for use with argon
arc welding”, British Welding Journal, Vol. 2, pp.471,
1955.
Yoshiki Mizutani, Takuya Tamura and Kenji Miwa,
“Microstructural refinement process of pure magnesium by
electromagnetic vibrations”, Materials Science and
Engineering A, 413-414, pp 205-210, 2005.
Y. Cui, C. Xu, and Q. Han, “Micro structure Improvement
in Weld Metal Using Ultrasonic Vibrations”, Journal of
Advanced Engineering Materials, vol. 9, pp. 161-163,
2007.
S.P. Tewari and Anand Shanker, “Effects of longitudinal
vibration on hardness of the weldments”, ISIJ international,
vol. 33, pp.1265-1269, 1993.
S.P. Tewari, “Effect of transverse oscillation on tensile
properties of mild steel weldments”, ISIJ international, vol.
39, pp. 570-574, 1999.
W. Weite, “Influence of vibration frequency on
solidification of weldments”, Scripta Materialia, vol. 42 ,
pp.661-665, 2000.
Lu Qinghua, Chen Ligong, and Ni Chunzhen, “Improving
welded valve quality by vibratory weld conditioning”,
Materials Science and Engineering A, vol. 457, pp. 246253, 2007.
B. Pučko and V. Gliha, “Charpy toughness of vibrated
microstructures”, Metallurgija, vol. 44, pp. 103-106, 2005.
INTERNATIONAL JOURNAL ON ADVANCED RESEARCH IN
SCIENCE, ENGINEERING AND TECHNOLOGY
(ISSN :
)
(A Quarterly Refereed Research Journal on Science, Engineering and Technology)
Information for Authors
1. All papers should be addressed to The Editor-in-Chief, International Journal on Advanced
Research in Science, Engineering and Technology (IJSET), Sri Krishna College of
Engineering and Technology, Kuniamuthur Post, Coimbatore : 641 008, Tamil Nadu,
India. E-mail : principal@skcet.ac.in and librarian@skcet.ac.in
2. Two copies of manuscript along with soft copy are to be sent.
3. A CD-ROM containing the text, figures and tables should separately be sent along with
the hard copies.
4. Submission of a manuscript implies that : (i) The work described has not been published
before; (ii) It is not under consideration for publication elsewhere.
5. Manuscript will be reviewed by experts in the corresponding research area, and their
recommendations will be communicated to the authors.
6. The manuscript should be in IEEE format
For publication and subscription details, please contact
The Librarian
Vankatram Learning Centre
Sri Krishna College of Engineering and Technology,
Kuniamuthur Post, Coimbatore : 641 008,
Tamil Nadu,
India.
E-Mail : librarian@skcet.ac.in
drfrlib@gmail.com
Mobile : +91-9750007196