You are on page 1of 9

# I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)

## Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com

Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 3 5

Abstract: This thesis introduces a new fast method for the
calculation of exponential B-splines sample at regular
intervals. As another approach, this paper presents an
exponential B-spline interpolation kernel using simple
mathematics based on Fourier approximation. A high signal
to noise ratio can be achieved because exponential B-spline
parameters can be set depending on the signal characteristics.
The analysis of these interpolated kernels shows they have
better performance in high and low frequency components as
compared to other conventional nearest neighbor, linear,
spline based methods. This new method is fast and it also
considered polynomial spline as special case. This algorithm
is based on a combination of FIR and IIR filters which
enables a fast decomposition and reconstruction of a signal.
For different values of the exponential parameter the
approximation function is obtained. In this thesis we have
tried to get the interpolation function which uses the
symmetric exponential functions of 4th order. When complex
values are selected for the parameters of the exponentials,
complex trigonometric functions are obtained. We are
considering the real part of these functions which is used for
interpolation of real signals corresponding to different
exponential parameter that leads to less band limited signals
when they are compared with polynomial B-spline
counterparts. These characteristics were verified with 1-D
and 2-D examples. We are also going through all the
interpolation methods which are already in use.
KEY WORDS: E-spline, Exponential B-spline,
Interpolation, medical imaging, X-rays.

1. INTRODUCTION
The problem of constructing a continuously defined
function from given discrete data is unavoidable
whenever one wishes to manipulate the data in a way that
requires information not included explicitly in the data.
In this age of digitization, it is not difficult to find
examples of applications where this problem occurs. the
relatively easiest and in many applications often most
desired approach to solve the problem is interpolation
where an approximating function is constructed in such a
way as to agree perfectly with he usually unknown
original functions at the given measurement points. In
view of its increasing relevance, it is only natural that the
subject of interpolation is receiving more and more
attention these days. Image expansion is required in
many facets of image processing. To generate precise
maps of the earths surface, cartographers, must expand
small regions of satellite image data. In medical imaging,
computerized tomography slices and X-rays may need to
be zoomed to search for anomalies. Reconnaissance
photographs must be expanded accurately to show hidden
details of weapons manufacturing plants and landing
strips. With common methods of image expansion the
distinguishing objects of such original images tend to be
smoothed over. This decreases the usefulness of the
expanded image in showing precise details.
The B-spline functions because of its close resemblance
with the sinc function were being started to use
prominently as an interpolation function. The term
spline is used to refer to a wide class of piecewise
polynomial function jointed at certain continuity points
called as knots. Until now, in the spline family,
extensive research is being done for polynomial spline
[Uns99a]. However, the exponential splines are more
general representation of these polynomial splines
[Dah87a]. In the present work, the continuous
exponential function is derived at equally spaced knots
using truncated power functions and for the formulation
of the exponential interpolated kernel this approximation
function is convolved with Fourier approximation of
the sampled exponential E-spline function [Leh99a]. The
calculation of polynomial B-splines is a particular case,
when the parameters of the exponents are set to be zero.
The exponential B-spline interpolation function is
derived for symmetric case taking different
exponential parameter in consideration.
A great variety of methods with confusing naming can be
found in the literature of 1970s and 1980s. B-splines
sometimes are referred to as cubic splines while cubic
interpolation is also known as cubic convolution, high
resolution spline interpolation and bi-cubic spline
interpolation. In 1983, parker, Kenyon and troxel
published the first paper entitled comparison of
interpolation methods followed by a similar study
presented by Mealand in 1988. However, previous work
of Hou and Andrews, as well as that of keys also compare
global and local interpolation methods. In more recent
papers, not only hardware implementations for linear
interpolation and fast algorithm and fast algorithms for
B-spline interpolation or special geometric transforms
have been published. However smoothing effects are most
bothersome if large magnifications are required. In
addition, shape based and objects based methods have
been established in medicine for slice interpolation of
three dimensional (3-D) data sets. In 1996, Apperdorn
presented a new approach to the interpolation of sampled
data. His interpolated functions are generated from a
Image Enhancement Method using E-spline

Ram Bichar Singh
1
, Anurag Jain
2
and Manoj Lipton
3

1
M.Tech Student, Radha Raman Institute of Technology & Science, Bhopal, India

2,3
Department of Computer Science, Radha Raman Institute of Technology & Science, Bhopal, India

I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 3 6

linear sum of a Gaussian function and their even
derivatives. Our work presents a comprehensive survey of
existing expansion methods.
Here, we propose E-spline method for image expansion
and compare to other methods such as Linear and cubic
spline. Many techniques currently exist for interpolation
and expansion. Commonly used methods are linear and
cubic spline expansion smooths the image data in
discontinuous regions, producing a large image which
appears rather blurry. Image interpolation has many
applications in computer vision. It is the first two basic
re-sampling steps and transforms a discrete matrix into a
continuous image. Subsequent sampling of this
intermediate result produces the resample discrete image.
Image expansion methods have occupied a peculiar
position in medical image processing. They are required
for image generation as well as image post processing. In
computer tomography or magnetic resonance imaging,
Image reconstruction requires interpolation to
approximate the discrete functions to be back projected
for inverse. The goal of this study was not to determine
overall best method but to present a comprehensive
catalogue of interpolation methods using E-spline, to
define general properties and requirements of E-spline
techniques.
Exponential spline plays a fundamental role in classical
system theory. During the past decade there has been
number of articles devoted to the use of polynomial
splines in image expansion. E-splines are a natural
extension of B-splines and have very similar properties.
B-splines are just a special case of E-splines (with
parameter alpha=0) these spline based algorithms have
been found to be quite advantageous for image processing
and medical imaging, especially in the context of high
quality interpolation where it has been demonstrated that
they yield the best cost quality tradeoff among all linear
techniques. The interest in these techniques grew after it
was shown that most classical spline fitting problems on a
uniform grid could be solved efficiently using recursive
digital filtering techniques. In continuous time signal and
system theory are the exponentials which plays a pivotal
role having made this observation and motivated by the
search for a unification between the continuous and
discrete time approaches to signal processing we decided
to undertake the task to find the parameter for image
expansion using Exponential splines. These splines, as
their name suggests are made up of exponential segments
that are connected together in smooth fashion. They form
a natural extension of the polynomial splines and have
been characterized mathematically in relatively general
terms. Even though there have not been many
computational applications of E-splines, we believe that
image expansion is one of the attractive and decent
applications of E-splines. The kinds of splines that are
most appropriate for signal processing are the cardinal
ones which are defined on uniform grid. It is proposed by
Michael Unser and Thierry Blu in their latest paper on E-
splines. Mathematically, this corresponds to the simplest
possible setup, which goes back to the pioneering work of
Schoenberg on polynomial splines in 1946. Since then,
there have been many theoretical advances and the
methods of spline construction have been extended for
non uniform grids and many other types of non
polynomial basis functions. The good news is that the
choice of a uniform grid leads to important
simplifications.
In the present study, we concern ourselves with matters
related to the computation of exponential splines. Our
primary goal is to develop new tension parameters
selection algorithms that expanded the image. It must be
emphasized that the lack heretofore of viable tension
parameter selection schemes has greatly diminished the
practical utility of exponential splines.

2. B-SPLINE APPROXIMATION
Basis splines (B-splines) are one of the most commonly
used family of spline functions. It can be derived by
several self convolution of a so called basis function.
Actually, the linear interpolant kernel can be considered
as the result of convolving the rectangular nearest
neighbor kernel so it is given as:
(1)
Therefore, Uniform B-splines can be obtained by
multifold convolution of rect functions:
(2)
We obtained the quadratic B-spline for N=3 which in
fact, equal the previously mentioned quadratic
approximation. Now, we obtain cubic B-spline for N=4
and is given as
:

(3)

The figure is shown below:

Fig 2.0.1 Kernel of cubic B-spline approximation

Note that the B-spline kernel fails to occupy the null
positions of the sinc function h4 (-1) =h4 (1) =1/6. And
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 3 7

therefore, the B-spline Kernel is actually not an
interpolation but rather than an approximation kernel.
2.1 Exponential B-spline Interpolation
To create an Interpolating B-spline kernel, the B-spline
approximation is applied to a different set of samples t
(k). Since the B-spline kernel is symmetrical and
separable, the reconstruction (1) yields

=
k
k x h k t x s ) ( ) ( ) (
(4)

With h=h4, as defined in (8). Note that the general case
(8) reduces to (1) if the samples are taken directly from
the image data: t (k) =s (k). Here, the t (k) must be
derived from the images sample points s (k) in such a
way that the resulting curve interpolates the discrete
image. From (8) and (7) we obtain
(5)
Which, ignoring edge effects, results in a set of
equations to solve

(6)
Labeling, the three Matrices above as S, C, and T
respectively. The coefficients in T may be evaluated by
multiplying the known data points S with the inverse of
the tri-diagonal matrix C.

(7)
In all other methods, the coefficients used for convolution
with the interpolation kernel are taken to be the data
samples themselves. Because the coefficients for B-spline
interpolation are determined by solving a tri-diagonal
matrix system, the resulting kernel Spline
h
(x).
For its simplification the interpolated image s(x) and the
data samples s (k) now are called u and v respectively.
From (10) we obtain v=t*c and in the frequency domain

(8)
Inversion of (8) yields
(9)
(10)
Hence (5) can be written as u=t*h4=v* Spline
h
(x) and
with (8) we finally obtain
(11)
The figure is shown below:

Fig 2.1.2(a) B-spline Fig 2.1.2(b) B-spline interpolated image
interpolation kernel

The interpolation kernel is symmetric, passes through the
integer points. Fig. 4 shows the interpolation kernel for
only [0,3], moreover the kernel is symmetric around x=0.

The Fourier domain response of the interpolation
kernel is shown in Fig. 4 for different values of
exponential parameters. As the value of alpha is changed
the filter response deviates from the ideal low pass
filter. However, the interpolation kernel is band limited
passing the high frequency components near the cut-off
frequency, which can be used to preserve the edge
information in the images. With the increase in value
of the transaction part of the filter decreases the
magnitude of the low frequency and increases the
number and magnitude of the high frequency
components.
Taking Fourier series approximation

(12)

(13)

(14)
Taking back in the spatial domain
Hence final equation:

(15)

) ) 2 cos( ( ) ( ) (
1
0

=
+ =
n
n
f a a f V f T
}
=
T
df f G
T
a
0
0
) (
1
}
=
T
n
df f f G
T
a
0
) 2 cos( ) (
1

=
+ + + =
1
0
))) ( ) ( ( ) (
2
( * ) ( ) (
n
n
n x n x a x
a
x v x t

=
+ + + =
1
0
))) ( ) ( ( ) (
2
( * ) ( ) (
n
n Espline
n x n x a x
a
x x h
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 3 8

(16)

(a) (b)
Fig 2.1.3 (a) Fourier domain magnitude plot and, (b) log
plot of Exponential B-spline for (, ,-, - ) with
different values of .

Table2.1.1The percentage energy distribution in
exponential B-spline interpolation kernel for different
after truncation.
Expone
ntial
parame
Energy
distributed
between -3 to 3
Energy
distributed
between -2
0 0.999800 0.998423
1 0.999941 0.998964
2 0.999991 0.999711
3 1.000000 0.999961
4 1.000000 0.999997

The energy distribution in exponential B-spline
interpolation kernel is minimum for lower value of
exponential parameter. As the value of these
parameter increases, energy decreases in both
between -2 to 2 and -3 to 3 which is shown in
Table1.
Fig. 2.4 plots the sum of sampled interpolation
kernel from equation (9) as a function of
displacement d. The summation is done after
truncating the kernel from -3 to 3. It is clear that
for alpha closer to 1.2, the sum of sampled
interpolated kernels is closer to 1, hence the value
close to 1.2 give better interpolation. This can be
verified with PSNR given in table 2.

Fig 2.1.4 Sum of sampled interpolation kernels as a
function of the displacement for different .

2.2 Cubic Interpolation:
Cubic polynomials are frequently used because of theory
ability to fit C2-continuous. Also, the B-spline
approximate h
4
as defined in (8) are constructed
piecewise from cubic polynomials. Of course, cubic
polynomials also can be used to approximate the sinc
function.
Two point Interpolation: in the case of cubic interpolation
with two points, a symmetric kernel can be defined with

(17)
The parameters A to D can be determined by applying the
following boundary conditions:
0 , 0 ) (
0 , 1 ) (
; ), ( ) (
; ), ( ) (
1
' '
0
= = -
= = -
= -
= -
+
+
k k h
k k h
continuity c k h k h
continuity c k h k h
(18)
For N=2, those boundary conditions yield four equations
for the four parameters resulting in;
(19)
It should be pointed out that, by definition, the above
cubic function is a DC-constant interpolator. The
resulting curves are similar to those obtained by linear
interpolation, but the pieces fit C1-continuously in the
spatial domain. Here, only DC-constant interpolators
have been derived in this subsection. The figure is shown
below:

Fig 2.2.5 Cubic interpolation=4

2.3 Gaussian Interpolation:
Appledorn has recently introduced a new approach to the
generation of interpolation kernels. The objective was to
exploit the characteristics of the Gaussian function in
both the spatial and the frequency domain. In particular,
the Gaussian function is recurrent with respect to
operations such as derivation and Fourier transform.
Hence, Appledorn published a scheme to develop simple
interpolation kernels that are both locally compact in the
signal space and almost band limited in the frequency
domain and in, addition are easy to manipulate
analytically.
Consequently, we will denote the Mth partial derivative
of the unit area Gaussian function
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 3 9

; with zero mean and
variance as :
(20)
Hence, we obtain

(21)

Then, the Mthorder Gaussian interpolation kernel is
given by
(22)
The weighting factors alpha and the variance are
determined from the following constraints:

- The Gaussian kernels should equal the ideal
interpolator, at least for x=0

- The Fourier transforms of the
Gaussian kernels should equal those of the ideal
interpolator, at least for

Furthermore, should be flat as possible
without any slope or curvature for
(23)
Note that the first and the second constraints cover only
one part of the interpolation condition (4) and the DC-
constant respectively. The latter constraint is imposed to
approximate the pass band characteristics of the ideal low
pass filter and therefore to minimize the corruption of the
images Fourier spectrum by the interpolation.
3. E -SPLINES
3.1 E-splines and its computations:
Exponential spline plays a fundamental role in classical
system theory. During the past decade there has been
number of articles devoted to the use of polynomial
splines in image expansion. E-splines are a natural
extension of B-splines and have very similar properties.
B-splines are just a special case of E-splines (with
parameter alpha=0) these spline based algorithms have
been found to be quite advantageous for image processing
and medical imaging, especially in the context of high
quality interpolation where it has been demonstrated that
they yield the best cost quality tradeoff among all linear
techniques. The interest in these techniques grew after it
was shown that most classical spline fitting problems on a
uniform grid could be solved efficiently using recursive
digital filtering techniques. In continuous time signal and
system theory are the exponentials which plays a pivotal
role having made this observation and motivated by the
search for a unification between the continuous and
discrete time approaches to signal processing we decided
to undertake the task to find the parameter for image
expansion using Exponential splines. These splines, as
their name suggests are made up of exponential segments
that are connected together in smooth fashion. They form
a natural extension of the polynomial splines and have
been characterized mathematically in relatively general
terms. Even though there have not been many
computational applications of E-splines, we believe that
image expansion is one of the attractive and decent
applications of E-splines. The kinds of splines that are
most appropriate for signal processing are the cardinal
ones which are defined on uniform grid. It is proposed by
Michael Unser and Thierry Blu in their latest paper on E-
splines. Mathematically, this corresponds to the simplest
possible setup, which goes back to the pioneering work of
Schoenberg on polynomial splines in 1946. Since then,
there have been many theoretical advances and the
methods of spline construction have been extended for
non uniform grids and many other types of non
polynomial basis functions. The good news is that the
choice of a uniform grid leads to important
simplifications.

In the present study, we concern ourselves with matters
related to the computation of exponential splines. Our
primary goal is to develop new tension parameters
selection algorithms that expanded the image. It must be
emphasized that the lack heretofore of viable tension
parameter selection schemes has greatly diminished the
practical utility of exponential splines.

3.2 Review of Theory:

A family of continuous piecewise basic spline functions
can be obtained by multifold convolution of functions
(24)
Where denotes the convolution integral operation.
These weights are defined in the
domain ] 2 / 1 , 2 / 1 [ e x , and being 0 otherwise for the
case of the centered basic splines and the
domain ] 1 , 0 [ x for the shifted ones. But in the case of
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 4 0

polynomial B-splines, (.)
i
w is a rectangular function
of height 1. In order to obtain the exponential B-splines,
the weights will be exponential functions
) exp( ) (
1
x x w
i
i i
w

with
C
i

being the
parameter of the exponential function and
C
w
i

1
a
normalization factor. So, by defining the
vector T
r
) .., .......... (
,

1

## the exponential B-spline can

also be expressed as:

(25)
Where the single and multiple discrete difference
functions are given by
(26)
Respectively, with
] [k
denoting the Kroneneckers
delta function. Also,
(.)

## is the continuous exponential

truncated power or exponential cone spline. The later can
be defined by

(27)

And then recursively calculated by

(28)

**Convention of sign: here =

In linear case for n=2 we have
T
) (
2 , 1 2
=

and from
above equation

(29)

(30)

And the truncated exponential can be simplified as
(31)
And therefore,
(32)
Hence,

(33)
(34)

3.3 Calculation of E-spline Basis Function:

A general solution of exponential spline function of order
n=4 is given below where the factors are defined as:

(35)
(36)

Here, we are considering for the symmetric case. By
putting the value of alpha1= , alpha2= , alpha3=-
, alpha4=- . The basis function which we get after
varying the value of alpha is given Table 1.

Table 3.2 Piecewise function for the E-spline
approximation of order n=4

Table 3.3.2
We see the asymmetrical behavior of spline approximat-
ion function from the above table and figure for different
order and are shown below:

Fig 3.3..6 (a) Nth order E-spline Fig 3.3.6 (b): Approximation
with varying with from function graph varying value
2 to 4 alpha
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 4 1

The graph for varying alpha in symmetric case is shown
above in fig 3.3(b)

3.4 E-spline Interpolation:
To create an interpolating E-spline kernel, the E-spline
kernel approximation is applied to a different set of
examples. Since we are dealing with symmetric E-spline
kernel the reconstruction yields:

=
k
k x h k t x s ) ( ) ( ) (
(37)
Here, h is approximation kernel which is discussed
above. The general case of given equation is reduces to
following equation

=
k l
D
l y k x h l k s y x s ) , ( ) , ( ) , (
2
(38)
If the samples are taken directly from the image data:
( ) ( ) t k s k .here the ( ) t k must be derived from the
images sample points ( ) s k in such a way that the
resulting curve interpolates the discrete image. From
equation 1 and 2 w e obtain

+
=
=
2
2
4
) ( ) ( ) (
k
k m
m k h m t k s

= ( 1) ( 1) (0) ( ) (1) ( 1) f t k f t k f t k + + + (39)

That results in a set of questions to solve when we ignore
edge effects:
(0) (0) (0) ( 1) 0
(1) (1) (1) (0) ( 1)
(2) (2) (1) (0)
( 2) ( 2)
( 1) ( 1) 0 (1) (0)
s t f f
s t f f f
s t f f
s K t K f
s K t K f f
( ( (
( ( (

( ( (
( ( ( =
( ( (

( ( (
( ( (

(40)

Labeling the three matrices above as S, C and T
respectively, the coefficients in T may be evaluated by
multiplying the known data points S with the inverse of
the tri diagonal matrixC .
1
T C S

= (41)
In all other methods included in this, the coefficients
used for convolution with the interpolation kernel are
taken to be the data samples themselves. Because the
coefficients for E-spline interpolation are determined by
solving a tri diagonal matrix system ( )
h
Espline x is
infinite.

To simplify its analytical derivation, the interpolated
image ( ) s x and the data samples ( ) s k now are called
u and v respectively. From we obtain * v t c = and in
the frequency domain

( ) ( ) *( (1) ( 1) (0) ( ) ( 1) ( 1))
( ) ( ) ( (0) 2 (1) cos(2 ))
v x t x f x f x f x
V f T f f f f

= + + +
= +
(42)
Inversion of above equation yields

( ) ( ) /( (0) 2 (1) cos(2 )) T f V f f f f = +

0
1 1
( ) ( cos 2 sin 2 )
n n
n n
v f a a nf b nf

= =
~ - + +

(43)
Here
n
b =0 and taking the inverse Fourier transform of
the above equation, the following equation will come:
0
1
( ) ( ) ( ) [ ( ) ( )]
2
n
n
a
t x v x a x x n x n

=
(
= + + + +
(

(44)
Here
n
b =0 and taking the inverse Fourier transform of
the above equation, the following equation will come:
0
1
( ) ( ) ( ) [ ( ) ( )]
2
n
n
a
t x v x a x x n x n

=
(
= + + + +
(

(45)
Hence, equation 1 can be written as
4
* *
h
u t h v Espline = = and with (13) we finally
obtain

( )
h
Espline x =
4
* ( ) h t x (46)
The graph is show below within the interval
3 x < .Although the kernel is infinite; the amplitudes of
the half waves are reduced significantly when compared
with that of ideal IIR-interpolation.

Fig 3.4.7 The interpolation Exponential B-spline kernel
with varying .

The Fourier response of the ( )
h
Espline x is given below
for varying the value of . Here solid line shows the
frequency response of =1. Subsequently, the dotted is
for =2. The following figure illustrated the side lobes
behavior of E-spline interpolation.

I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 4 2

Fig 3.4.8 4
th
order Exponential B-spline approximation and
normalized Exponential B-spline for ( , ,- ,- ) where
=.2,.5,1,2,3

Here we are considering the case for n=4( , , - ,-
). The following figure shows the impulse response in
spatial domain of 4
th
order , , - , . Cardinal
exponential B-spline decays faster than polynomial spline
and also we see that increase the value of ranging
from .2 to 3. The ringing behavior is reduced. The second
figure shows the logarithmic plot of Fourier magnitude
which shows energy distribution for different frequencies
in the main lobe and side lobes. The energy distribution
in the E-spline is minimal for lower value of alpha and as
the values of alpha increases the energy decreases in both
between from -2 to 2 and -3 to 3 which is shown in table
3.4.3.

Exponential
parameter
Energy between
-3 to 3
Energy in other
part
0.2 99.9800 0.0112
1 99.9941 0.0059
2 99.9993 0.00085739
3 100 0
4 100 0

Exponential
parameter
Energy between
-2 to 2
Energy in other
part
0.2 99.8431 0.1578
1 99.8970 0.1036
2 99.9973 0.0289
3 99.9965 0.0038
4 99.9997 0.00030095
Table 3.4.3

The given image is the expansion by a factor 8 by using
the above E-spline technique in 2-D. Here, we are
performing E-spline interpolation on Lena image
(64x64). Lena image is expanded by factor 8. Here are the
given results of varying the value of .

Fig 3.4.9 (a) Original Image Fig 3.4.9 (b) =0.2

Fig 3.4.9 (c) =0.5 Fig 3.4.9 (d) =1

Fig 3.4.9 (e) =1.5 Fig 3.4.9 (f) =2

Fig 3.4.5 Expansion of Lena image by Exponential spline
with varying
4. EXPERIMENTAL RESULTS
Here we perform on lena image (64x64). The given image
is the expansion by factor of 8 from a section of the image
of Lena. After performing, E-spline interpolation with
varying . We get the smooth area corresponding to the
face of Lena and higher frequency parts which are the
back ground. In above figure it can be observed that the
resulting interpolated images are almost the same at a
glance, but more jaggedness appears in the high
frequency areas in the exponential B-spline case due to its
fewer bandslimited characteristics. The Fourier analysis
of pass band, stop band, and cutoff frequency the nearest
neighbor and linear interpolation should be avoided while
the preferred method is the Gaussian kernel with large
sizes. To judges the interpolation quality, it solely
depends on the application. Interpolation versus
Approximation: in this E-spline methods are well suited
for the images containing high frequency components.
Gaussian kernels are not suited for this method. These
kernels have been compared on various images of Lena.
In each case, the efficiency and accuracy of a particular
interpolation technique was evaluated by analyzing its
Fourier properties, visual quality and run time
measurement.

4.1 Runtime measurements:
The runtime of the various interpolation schemes were
measured on the standard machine. Sources have been
compiled using MATLAB 6.5. The rotation is quite time
consuming in MATLAB environment. It shows that
simple interpolation methods such as nearest neighbor
,linear and 2x2 cubic interpolation are fairly fast and
requires less time than the rotations of pixel coordinates.
I nt ernat i onal Journal of E mergi ng Trends & Technol ogy i n Comput er Sci ence (I JE TTCS)
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com
Volume 1, Issue 4, November December 2012 ISSN 2278-6856

Vol ume 1 , I ssue 4 November - December 2 0 1 2 Page 4 3

Gaussian interpolation required more time due to the
evaluation of the exponential function necessary to
determine weights. Here are the results of all interpolated
image and we compare all interpolated image to original
one. The resultant image is show below:

Fig 4.1.10 (a) Original image Fig 4.1.10(b) Nearest
neighbor

Fig 4.1.10(c) Linear image Fig 4.1.10(d) B-spline
image
Fig 4.1.10 Interpolated image of Lena (64x64) using
various types of interpolation methods.

5. CONCLUSIONS
This thesis is focused on interpolation methods using E-
spline. The method is fast for calculation of E-spline
(where the calculation of B-spline is a particular case).
When complex parameters are used in exponential
functions, trigonometric spline are obtained and here we
are dealing with only real parts of the signal. Less band
limited functions can be achieved by using the real par of
this later functions and comparing them with the
polynomial splines counterparts. Although more amount
of computation is required, it is specially used for images
containing high frequency components. As demonstrated
by codes in this thesis, the interpolation methods using E-
spline is easy to implement. The E-spline is general case
of B-spline polynomials and performs better interpolation
for images containing high frequency parts.
6. FUTURE WORK
Since I have a good result for the interpolation using E-
spline techniques, the further study on interpolation
should be carry on in the future. Interpolation of images
is popular problem until yet but now we have to find
kernel that enhances high frequency parts. If we get the
required result then we go for L-spline that is general
case of E-spline. Future work will address techniques to
get better result in interpolation methods by finding the
adaptive method for the appropriate image.

REFERENCES
 M.Unser, Splines: A perfect fit for signal and
image processing, IEEE Signal Processing
Magazine, Vol.16, No.6, pp.22-38,November 1999.
 Raghubansh B. gupta, Byung Gook Lee, Joon Jae
Lee: A new Image Interpolation Technique using
Exponenial B-Spline.
 Thomas M.Lehmann, Claudia Gonner, and Klaus
Spitzer, Survey: Interpolation Methods in Medical
Image Processing, IEEE Transactions on Medical
Imaging, Vol. 18, No. 11, November 1999.
 Takeshi Asahi, Koichi Ichige,and Rokuya Ishlii,
Fast Computation of Exponential Splines
Proceedings of the IEEE.
 M.Unser,A.Aldroubi and M.Eden, Fast B-spline
transforms for continuous image representation and
interpolation, IEEE Trans. On Pattern Anal. &
Machine Intell. Vol.13, No.3, pp.277-285, March
1991.
 Erik Meijering,A Chronology of Interpolation:
From Ancient Astronomy to Modern Signal and
Image Processing, Proceedings of the IEEE, vol.90,
No.3, March 2002.
 Michael Unser and Thierry Blu, Cardinal
Exponential Splines: Part I- Theory and Filtering
Algorithms, IEEE Transactions on Signal
Processing, Vol. 53, No.4, April 2005.
 Michael Unser, Cardinal Exponential Splines: Part
II- Think Analog, Act Digital, IEEE Transactions
on Signal Processing, Vol. 53, No.4, April 2005.
 Brian J.Mcartin,Compuatation of Exponential
Splines SIAM J. Sci STAT. Comput. Vol. 11, No. 2,
pp.242-262, March 1990.
 Poth Miklos,Image Interpolation techniques.