950 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5 , NO.

6, JUNE 1996

A General Framework for Quadratic

Volterra Filters for Edge Enhancement
Stefan Thurnhofer, Member, IEEE, and Sanjit K. Mitra, Fellow, IEEE

Abstruct- An inherent problem in most image enhancement

schemes is the amplification of noise, which, due to Weber’s law, is
mostly visible in the darker portions of an image. Using a special
class of quadratic Volterra filters, we can adapt the enhancement I I
edge
process in a computationally efficient way to the local image
extraction
brightness because these filters are approximately equivalent to
the product of a local mean estimator and a highpass filter. We Fig. 1. Block diagram of the unsharp masking technique.
analyze and derive this subclass of quadratic Volterra filters by
investigating the 1-D case first, and then we generalize the results
to two dimensions. An important property of these filters is that 0 -1 0 -1 -1 -1 1 -2 1
they map sinusoidal inputs to constant outputs, which allows us -1 -1 -1 -1 -2 -2
to develop a new filter characterization that is more intuitive for 0 -1 0 -1 -1 -1 1 -2 1
our application than the 4-D frequency response. This description (Laplacian I) (Laplacian 11) (Laplacian 111)
finally leads to a novel least-squares design methodology. Image
enhancement results using our Volterra filters are superior to Fig. 2. Filter kernels for three Laplacian operators
those obtained with standard linear filters, which we demonstrate
both quantitatively and qualitatively.
true when images are not perfectly clean but were sent through
transmission channels or were otherwise corrupted since noise
I. INTRODUCTION in dark areas is very visible, and it should not be amplified
even more. For example, consider the structure of Fig. 1. The
HE human visual system is able to discern details in im-
input image is sent through a block that extracts edges and
ages depending on the characteristics of the surrounding
features. The output is then scaled by an appropriate factor p
area. Many different effects can influence this process, and
and added back to the original image. This method is generally
various types of masking are described in the literature [l].
referred to as unsharp masking [2],[3] and is quite effective
One of the most important ones is commonly known as the
for enhancing low contrast images. The edge extraction block
Weber-Fechner law, and it applies in a similar form to other
in Fig. 1 is often implemented as a linear highpass filter
sensory processes as well. It characterizes the perception of
such as a discrete Laplacian operator for which three possible
grey-level differences in dependence on the surrounding back-
filter kernels are shown in Fig. 2 [ 3 ] .The main advantage of
ground image intensity. Roughly stated, the just noticeable
using this particular filter is its computational simplicity. The
difference (JND) is proportional to the average intensity of the
highpass filter enhances those portions of the image that bear
surrounding pixels. In other words, we can see details more
mostly high-frequency information, i.e., edges and textured
easily in dark regions, whereas bright portions tend to mask
regions. The perceptual impression is improved because the
details. Even though Weber’s law does not describe such a
image appears sharper and better defined.
dependence for very small intensity values, we can still use
An apparent problem of this technique is that it does not
the proportionality as a good overall estimate.
discriminate between actual image information and noise.
As a direct consequence of Weber’s law, image processing
Thus, noise is enhanced as well. To eliminate this problem
systems (image enhancement, image coding, image restoration,
while still preserving the simplicity of the algorithm, we make
etc.) must treat the darker regions of an image very carefully
use of the property of the HVS described by Weber’s law
because a human observer would perceive any imperfections
and modify the method such that the image enhancement
more easily here than in the brighter areas. This is especially
is dependent on the local average pixel intensity. In bright
regions, we can enhance the image more because noise and
Manuscript received December 15, 1994; revised December 11, 1995. This
work was supported by the SDIORST, managed by the Office of Naval other gray-level fluctuations are much less visible. On the other
Rescarch under contract ONR N00014-85-K-0551 and by a University o f hand, in darker portions, we want to suppress the enhancement
California MICRO grant with matching funds from Hughes Aircraft Co., process since it might deteriorate image quality. This simple
Xerox Corporation, and Signal Technology, Inc.
S. Thurnhofer is with Lucent Technologies, Bell Laboratories, Allentown,
idea indicates the need for a highpass filter that depends on
PA 18103 USA (e-mail: stefan.thumhofer@bell-labscom). the local mean similar to
S. K. Mitra is with the Department o f Electrical and Computer Engineering,
University of California, Santa Barbara, CA 93106 USA.
Publisher Item Identifier S 1057-7149(96)04184-X. H ( w ) cx H ~ ( w. )(local mean) (1)
1057-7149/96$05.00 0 1996 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 95 1

where HtL(w)is a linear highpass filter. The most obvious We can then write
solution of this problem is to multiply H h ( w ) with a linear
local mean estimator (i.e., an averaging filter). The exact y(n) = hz(n,1,n2)* *2(n>1,n2)l
n=n1=n2 (4)
computational expense for such a structure depends on the
particular filter design, of course. For a filter with a region where ** denotes 2-D convolution. For example, consider the
of support of 3 x 3 pixels, we have to expect abjout 10 system given by ~ ( n=) 2z(n - 1)x(n) - 3 ~ ( n ) x~( n + +
multiplications and 16 additions. We will see that using l)z(n - 2 ) . Using the notation for unit-sample sequences
quadratic Volterra filters, a much more efficient solution can
be found that only needs seven mulliplications, of which two
could be implemented as shift operations and four additions.
Among the various forms of nonlinear filters, Volterra filters the Volterra kernel for this system is then hZ(nl,n2) =
represent the most natural extension of linear filters. Volterra 26(nl - 1,1%2) - 36(nl,n2) + 6(TLl + 1,1%2
2). -

filters can be described as essentially a linear filter with higher Even though Volterra systems are nonlinear in the input se-
order polynomial extensions. Thus, reducing a Volterra filter quence, they are linear in the kernels. It is this observation that
to its first-order component yields a linear filter. On the allows us to use frequency domain techniques for the analysis
other hand, this relationship with linear filters is tht. basis and design of these systems. We define the spectrum of the
for some very useful properties of the higher order terms. Volterra kernels as the multidimensional Fourier transform of
For example, we can define the describing parameteIs as a h,2(ni,n2) [SI, 161
generalized impulse response and its multidimensional Fourier
transform (or Laplace transform for the continuous-time case)
as a generalized frequency response. Even though the filter
is not linear with respect to the input signal anymore, it is nl=-w nz=-oo
still linear in the impulse response coefficients, i.e., a linear
combination of filters is equivalent to a filter with the same The symbol .F denotes the Fourier transform in one or more
linear combination of the kernel parameters. Another important dimensions, depending on the context. We call H z ( w l ,w 2 )
observation is that Volterra filters work on products of input the second-order generalized frequency response [6] or,
samples. Both of these simple properties are fundamental for for simplicity, second-order frequency response. Similarly,
the analysis because it allows us to write a Volterra filter as a hz(n1.n ~ is) called the second-order impulse response.
multidimensional convolution operation. Similar to the design problem of linear filters, the frequency
We restrict ourselves to second-order filters only since they domain representation is an important tool for the design of
are the smallest-degree filter that exhibit a nonlinear behavior. Volterra filters. We rewrite (4) using the frequency responses
Higher-order filters can potentially be very expensive in terms of input and system as
of number of computations, and for implementation purposes,
it is desirable to keep the number of arithmetic operations as (8)
V(n) = - T - 1 { H L ( w l . w Z ) x ( w 1 , w 2 ) } 1n=n1=n2
small as possible.
In Section 11, we describe the basic properties and rellation- where X ( w ) = . F { x ( n ) }and X ( w 1 , w z ) = F { % ( n l , n z ) =
}
ships of quadratic Volterra filters. Section I11 defines various X ( w l ) X ( w 2 ) .Unless noted otherwise, we use lowercase let-
subclasses of quadratic Volterra filters, which can be vvritten ters for sequences in time or space and uppercase letters for
as mean-weighted highpass filters in one and two dimensions. their Fourier transforms.
We also introduce a scheme to describe their dependence on Unlike for linear filters, however, the impulse response of
input frequency and orientation. Based on this description, we quadratic Volterra filters does not necessarily define the filter
use a least-squares approximation to design a 2-D filter for uniquely, i.e., there can exist many kernels for the same filter
perceptual edge extraction. Finally, Section IV illustrates the [7]. Only if the kernel is symmetric does it uniquely describe
application to image enhancement. the system, and for a second-order system, this means that [8]

G(n1, n2) = q n 2 , n l ) (9)

11. QUADRATIC
VOLTERRA FILTERS:
which we will use frequently. Since every kernel can be
DESCRIPTION
AND RELATIONSHIPS
symmetrized [SI, (9) does not impose any restrictions.
A discrete-time quadratic Volterra filter is defined by 141 The 2-ID Volterra series is a straightforward extension of
00 00
(2) [9]. Here, the input and output signals are 2-D sequences
(usually called “images”). We write the 2-D discrete-time
Y).( = V2(.) = hz(kl,kz)z(n-kl)z(n-k.2).
kl=--oo &=-a
quadratic Volterra filter as
(2)
This expression can be interpreted a.s a 2-D convolution of
hz(n1,n2) with products of samples of the input sequence.
This becomes clearer if we define

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 952 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5, NO. 6, JUNE 1996

with 3(nJ1,n2,n3,nq) = z(n1,n ~ ) z ( n n 3 g, ) . We assume that if

symbols like zor y can represent both 1-D and 2-D sequences,
depending on the context. In (lo), we have omitted the limits
of the sums. In the most general case, all variables kl,. . . ICg
will range from -cc to 00. All the filters that we consider
here are of finite extent, however. If, for instance, the impulse
response of the second-order term is such that the filter only
uses pixels that are a maximum of N I pixels away from
the current pixel in n,l direction and N2 pixels away in n2
direction, then - N I 5 k l , k3 IN I , and -Nz 5 k z , kq 5 N2.
In general, the limits are of minor importance, and in the
following, we will omit them for simplicity unless their values
cannot be inferred easily.
Using again the Fourier transform notation, we can rewrite
(11) as
The approximation error is then given by

where X ( w ) = X ( W ) - 27rpL,S(w) is the mean-removed

version of X ( w ) .
Pro08 The proof of this theorem is given in [12]. The
basic idea is to decompose the general expression of quadratic
HIGHPASSFILTERS
111. MEAN-WEIGHTED
Volterra filters given in (2) into the two terms
We now define a subclass of quadratic Volterra filters
and derive its properties. The filters from this class are
approximately equal to a product of a local mean estimator
and a highpass filter. We investigate the 1-D case first since
the arithmetic expressions are simpler than those for the 2-D
ki ka
case.
where ? ( n )= z ( n )- p x is the mean-removed version of z(Tr1).
A. I - D Mean- Weighted Highpass Filters We then show under which conditions yz(n) can be neglected
compared with yl(n). 0
In [lo], a special quadratic Volterra filter called Teager's
We need to make some remarks about this theorem. First,
algorithm has been defined, and in [11], it has been shown
(16) is true for practically all images. In those cases where
that this filter is approximately equal to a product of local
it should not be true, the signal or image would have to be
mean and highpass
scaled and shifted accordingly before processing. Second, we
can approximate the mean p, in (15) by a local averaging filter
y(n) = ."n) - z(n - l)z(n + 1) (13) if we assume that the kernel of the Volterra filter is finite. In
k,(Zz(n) - z ( n - 1) - z ( n + 1)) (14) this case, only pixels in a local neighborhood influence the
result, and thus, only the local characteristics of the signal
+ + +
where k , = ( z ( n- I) z ( n ) x(n 1))/3 estimates the play a role for the computation of the output. Third, there is no
local mean. Additionally, 2-D extensions of the Teager filter restriction on the spectral shape of Hz(w,0). For applications
have been introduced in [ l l ] , which were found intuitively in edge enhancement, however, we will require highpass
and could be decomposed analogous to (14). These filters have characteristics, in which case, the corresponding Volterra filter
been used successfully for image enhancement. However, in is a mean-weighted highpass filter. For the most general case,
[l I], it was not clear whether or not these filters would be we can relax the conditions in (16) and (19) even further and
optimal in any way or what is the general class of Volterra drop the symmetry condition in (18), but then, we need to
filters to which they belong. We can now answer the latter rewrite (15) as
question for 1-D filters with the following theorem, where
we model the input sequence z ( n ) by an i.i.d. process with
uniform distribution and with mean px and standard deviation
a,. Even though (19) appears fairly complicated, we can greatly
Theorem: A 1-D second-order Volterra filter can be approx- simplify it for certain special cases. In the first class, filters
imated by a mean-weighted filter, i.e. consist of squares of pixels.
Dejnition: If h~(n1,na) = 0 for all n1 # nz and
E,, E,, hz(n1,nz) = 0 and Hz(w,O)has highpass

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRP,TIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 953

and thus

The locations of the nonzero coefficients of the impulse

response are indicated in Fig. 3(b). These filters have the
additional useful property that they are symmetric about the
current sample. The symmetrized Teager filter y(n) = x 2 ( n )-
+ +
0.5z(n- l ) z ( n 1)- 0.5x(n 1)x(n- 1) is an example for
a simple system from this class [12].
The Teager filter maps sinusoidal inputs to constant outputs
(a) (b) [lo], and it is straightforward to show that, in fact, every
Fig. 3. Locations of nonzero impulse response coefficients of special cases class I1 system has this property. Using (21), we can see that
of mean-weighted highpass filters: (a) Class I filters; (b) class I1 filter.
H z ( w , w ) = C h 2 ( k , - k ) = H 2 ( 0 , 0 ) , which is assumed to
+ +
be zero. With X ( w ) = -j7rS(w - W O ) j n S ( w wg) and the
characteristics, then the filter is called a class I discrete-time spectrum of the output signal given by [12]
quadratic Volterra filter, or class I filter for short.
The following lemma states the approximate behalvior of
class I filters.
Lemma I : Class I filters can be approximated as ZL mean- the output consists only of a dc component.
weighted highpass filter.
Proof: Equation (17) holds by assumption, and (18) is B. 2 - 0 Mean- Weighted Highpass Filters
true since we have hz(n1,na) = Cy=-, h 2 ( k , k ) 6 ( n l - We now consider the extension of the previous results to 2-
k , n 2 - k ) , which leads to D signals for applications in image processing. Since the class
00
I1 filters of Lemma 2 inherently lead to symmetric systems,
Hz(w1, w2) = h 2 ( k ,k ) e - + - J k W z the localization of edges is preserved when we apply class I1
f=-00 filters to images. This property is especially important here
and thus H z ( w , 0) = h z ( k , k ) e - J k w = Hz(0, w ) . because otherwise, it would be impossible to enhance edges
It remains to show that S h 2 0. In (19), all the sums since the exact location of the feature would be unknown.
yield zero because h z ( k l , k2) is nonzero only for kl = k2. The fundamental building block of class I1 systems is
Therefore, we obtain Sh = 0 here. 0 z(n - k)x(n +
k ) . The overall system consists of a linear
Examples for this class are y(n) = z 2 ( n )- 0.5z2(n - 1) - combination of several of these terms for different k . Thus,
0.5x2(n-3) and y(n) = z 2 ( n - 1 ) - - x 2 ( n - 2 ) . The first filter each term is a product of samples that are equally far away to
can approximately be written as p z ( 2 x ( n )- x ( n - 1)-- z ( n- the left and right of the center location n. We extend this idea
3)) and the second as 2 p z ( x ( n - 1) - x(n - 2)). Fig. 3(a) to the 2-D case simply by defining the basic building block of
shows the possible locations for nonzero coefficients of the the 2-D filter as z(n1- k l , nz - k z ) z ( n l + k l , 722 k z ) , i.e., it +
impulse response in the nl-n2-plane. consists of pixels centered around the current pixel z(n1,n2).
The second family of filters is more important for our For the impulse response, we obtain the simple property
applications. h , 2 ( n l l n 2 , n 3 , n 4 )# 0 only if n1 = -n3 and n2 = -724.
Definition: If h 2 ( n l r n 2 ) = 0 for all n1 # -712 and
(23)
E,, E,, h2(a1, 7L2) = 0 and H2(w, 0) has highpass char- Analogous to Lemma 2, we also require that
acteristics, then the filter is called a class I1 discrete-time
quadratic Volterra filtei, or class I1 filter for short.
The following lemma states the approximate behalvior of k.1 k.2 k3 f4

class I1 filters.
Additionally, we can show that
Lemma 2: A Class I1 filter can be approximated as a mean-
weighted highpass filter.
Proof: Here, we can again simplify S h , and with the
symmetry condition for hz ( k l , k z ) , we obtain To prove this, we write the kemel as

S" = x h ; ( k , - k ) 2 0.
k
k#O

In a similar fashion as in the proof of Lemma 1, we can

show that Then, we find that
00

k=l

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 954 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5 , NO. 6, JUNE 1996

and thus a=- 1

a<O a<O

where we have used the symmetry of the kernel, i.e.

ko t
a=l
b>O
b=O
Using (23)-(25), we find the corresponding extension to
Fig. 4. Parameters a and b determine the orientation of the input sinusoid.
(15) to be They are interlinked by (I' +
h' = 1.

(i.e., arbitrary images) remains predictable. This simplification

allows us to describe the behavior in more intuitive terms than
Assuming that Hz(w1. w2,0,0) has highpass characteristics in by using the complete frequency response.
either the w1 or w2 direction or in both, we conclude that the
The choice of the input is very much determined by our
class I1 2-D systems can be approximated by mean-weighted
design goals. The signal must be such that the system response
2-D highpass filters.
is meaningful so that we can draw conclusions as to the extent
to which we have achieved the desired behavior. To this end,
C. Characterization of Edge Extracting Filters we choose a single sinusoidal signal as the system input. By
varying its frequency, we can roughly estimate how the system
We have developed the theory behind the class of filters that will respond to more complex input signals with arbitrary
we want to use for edge extraction. We know their structure frequency content. Even though it will not be possible to
and properties in both space and frequency domains, and we predict quantitatively how a system will react to an arbitrary
also have justified why, at least in principle, this family would input signal, because we do not take into consideration any
be advantageous for our goal. In the next step, we need to intermodulation frequencies or harmonics, we still can make
develop a performance measure that somehow allows us 1.0 find qualitative assertions. If the region of support of the filter is
the best filter from this class. Even though essentially all mean small enough, we can model the corresponding portion of the
weighted highpass filters would extract edges of an image in input image by a single 2-D sinusoid. The approximation error
some way, there are still important differences that we have will decrease with the size of the region of support (typically
to take into account. For instance, a problem that would not
5 x 5 or 3 x 3).
occur in I-D signal processing but is of extreme importance for As we have already indicated at the beginning of this
image processing is isotropy. Of course, the filter should find
section, isotropical behavior is of critical importance. To
edges independent of their orientation. Horizontal or viertical
incorporate a test for isotropy, we use rotated sinusoids with
boundaries must lead to the same response, and therefore, the
the rotation as a free parameter
degree to which a filter is isotropic is critical for the resulting
image quality.
In this section, we introduce a method of characterizing the
isotropical behavior as well as the dependence of a filter output
on the input frequency [13]. Unlike for linear filters, however,
where a and b determine the orientation, and ' a +
b2 = 1.
In Fig. 4, the possible orientations are indicated with the
input frequencies have a rather complicated influence on the
corresponding values for these parameters. Both must be
output, and we cannot show this dependence in any cornplete
between -1 and 1, and because they are interlinked, only
and yet easily interpretable way. Even though the filter is
one of them is sufficient to define the orientation. In addition,
mathematically completely defined by the frequency response
for our purposes, half of the possible directions in Fig. 4 are
H2 (.), a plot of this function does not usually yield any insight
redundant, and we will only use
into the edge extracting capabilities of the system. Essen tially,
two factors make the analysis hard to interpret: the complexity
of the system and the complexity of the input signal. Even for
simple quadratic systems, the frequency domain representation h = 2/1 - a 2 .
can appear confusing. Above all, we cannot restrict our design
to systems for which we have a simple intuitive explanation. The output spectrum of a 2-D quadratic Volterra filter is
This leaves us with the second source of complexity: the input given by [12]
signal. Of course, we have complete control over this for
design purposes, which means that we can develop a filter that
reacts in a specific way to a certain input. However, we have
to make sure that the filter's response to real-world signals

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 955

We find The corresponding frequency responses are

= (1 - cos(w1
H;a)(W1,W2,Wj,w4) - Wj))

x (1 - cos(w2 - wg)) (35)

Y y ( a ) ( w O ,U ) = 2 7 4 1 - cos(woa))
x (1 - cos(wob)). (36)
Fig. 5(a) shows the isotropy plot. The right-hand plot displays
the same graph viewed from the side. Here, the U axis is no
longer visible since it is perpendicular to the paper plane with
the positive part pointing into the paper. This side view makes
As expected, the output consists of components with zero and the dispersion clearly visible, i.e., we can assess the extent to
twice the input frequency. We can, however, easily show that which the filter varies its response for different values of U .
only the dc component will be nonzero. We use (26) and (23) We see that Filter A does not respond at all for 0, = -1,
and obtain 0 and 1, corresponding to horizontal and vertical orientations
(see Fig. 4). Similarly, the image in Fig. 6(a) shows the strong
-h, - k z ) == 0. directional dependence of the filter output. We can make
another observation by examining the way the response rises
for increasing frequencies in Fig. 5(a). For small values of
Thus, only the terms with S(wl,wz) remain in (31). This is
W O , the output is very small, and it starts to rise only for W O
not too surprising because the class I1 2-D filters were derived
approximately greater than 0.5 rad. This means that the system
as an extension of their 1-D counterparts in Section 111-A and
has a strong highpass characteristic. Accordingly, the lines in
for those we have already shown this property. Thus, we write
Fig. 6(a) are very thin.
In [ 111, the authors introduced heuristically found ex-
tensions of Teager's algorithm. Now, we can study their
properties in the framework of the isotropy plots. The first
with ~i'~)(wg, a ) = 7 r 2 [ ~ 2 ( w o aweb,
, - q a , -web) +
of these systems (Filter B) is an application of the Teager
Hz(-woa, -web, w o ~ , , w o b ) ] . filter along diagonals.
Summarizing, we have developed a method that allows us
to characterize the behavior of class I1 filters in an intuitive db)(71.1, 71.2)

way. Only three values play a role in this scheme: the input = 2zy71.1, 712) - z(n1 1.712 1) + +
frequency, the orientation of the input sinusoid, and the level x z(n1 - 1 , 7 1 2 - 1) - z(n1 1,712 - 1) +
of the constant output. Most importantly, we can show the
relationships graphically, which is a fundamental step toward
x x(n1 - 1,712 1) + (37)
intuitive understanding. Our assumptions greatly simplify the H;"@l, w2. wx,w4)
originally rather complex situation. = 2 - cos(w1+ w2 - wj - w4)
We now want to illustrate how to interpret the plots of - cos(w1 - wz - wg wg) + (38)
directional and frequency dependence. We call these graphs
isotropy plots because they indicate to which degree i2 filter Y;c(b) (WO, a)

is isotropical. In order to understand how these properties = 2 7 4 2 - cos(2woa 2 w o J I - a 2 ) +

determine the system output for images, we also use an - COS(2W"U - 2w02/1 - U * ) ) . (39)
artificial image shown in Fig. 6(a). It is a ring with added
Gaussian noise ( 0 2 = 0.01). The image dimensions are Fig. 5(b) shows that this system is fairly isotropic up to about
256 x 256 pixels, and the inner and outer radii of the ring are 80 wo = 0.5, which we can verify visually in Fig. 6(b).
and 100 pixels, respectively. Before adding the noise, pixels on Filter C is similar to Filter B, but along the horizontal and
the ring have unity value, and outside, they are zero. The step vertical directions, and it is defined by
edge is almost ideal, thus containing very high frequencies,
which will illustrate the differences between the filters more
y(C)(n1, n 2 ) 2x2(711,n z ) - z(n,1 1,n 2 ) +
x z(71.1 - 1,'TLz) - Z(71.117L2- 1)
clearly than a smooth transition.
We define the first example system (Filter A) by the input- x 5(7L1. r12 1) + (40)
output relation H$')(wl,w2,"-'3, '"4)
= 2 - COS(W~- ~ g -) C O S ( W Z - wg) (41)
Y;c(c)(wo,
U) = 27r2(2 - cos(2woa)
cos(2wo2/1 - a"). (42)

The isotropy plot in Fig. 5(c) appears even smoother than

for Filter B (see (37)) and shows isotropical behavior up to
higher frequencies. In addition, Filter C has a more pronounced

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 956 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5, NO. 6, JUNE 1996

w
....... ..; .... ....................... . . . . . . ~

3 25

............. ..-.
............................. :. . . .

Fig. 5. Isotropy plots for the example filters. The left graph shows the regular view, and the right one displays it from the side with the a-axis
perpendicular the paper plane.

highpass characteristics, which explains why the lines in

Fig. 6(b) are thicker than in Fig. 6(c).
with ckl Ck2hz(k.1,k 2 , - k l , -kg) = 0. Obviously, any
filter from this class is a linear combination of terms like
Summarizing, we find that for the design of an edge + +
z(nl - k l , ri2 - ~ Z ) Z ( T L ~ kl,ng k2). This is based on the
extracting filter, the first system would not be a good choice property of Volterra filters that they are linear in the kemels.
because of its very unisotropical properties. Either Filter B or We make use of this observation and introduce the concept of
C would be better options, where Filter C is slightly more basis filters for the design of Volterra filters.
sensitive to details. We define an expression

D. Least-Squares Design yicl k 2 (711.712) = x2(n1,112) - x(n1 + kl, + k z )

n2
By combining (10) with (23), we obtain the input-output x z(n1 - k l , rL2 - k2) (44)
relation for class I1 filters [14]
as a basis for class I1 filters, where 0 5 k1 < m, -00 < k 2 <
Y(”1.122) = J L Z ( h , k . 2 , -k1, - k . 2 ) 00, and k l +1k.21 # 0, i.e., kl and kg cannot be equal to zero
kl kz at the same time. We only consider unique pairs of kl and kz.
z ( n l - Icl) n2 - k2)x(nl + ICl n2 + k 2 )
1 If we would also permit ICl < 0, then some of the filters would
(43) be identical because ykl,lc2(n1,n g )= y--lcl,-kz(n1,n2). Using

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 957

Fig. 6. Original “ring” image and results of applying the example filters to it.

(44), we find the equivalent expression to (43) as this set. Furthermore, the basis filters have the advantage that
for each of them the sum of their coefficients equals zero, and
Y ( n l > n 2 )= ~ ~ B k l , k Z / B l , k 2 ( ~ l , ~ 2 (45)
) thus, (17) holds by design, and the coefficients i9kl , k 2 can have
kl kz arbitrary values. The relation between the kernel coefficients
and the t?kl,kL is
where the O k l , k z represent the design coefficients. We prove
this equivalence by rewriting (43) as
h z ( O , O , 0,O) = 7; k , k 2 (46)
ki kz
JL2(n1,nz1-n1, -4 = - @ n l , n z (47)
for nl + In21 #0
O<nl<oo -oo<nz<oo
h 2 ( n 1 , n 2 , -RI, -n2) =0 for nl < 0. (48)

In the frequency domain, the linear dependence on the basis

ki kz filters is preserved by the Fourier transform. Since we know the
frequency characteristics of each of the basis filters, we must
find the coefficients in such a way that the overall system has
a certain desired frequency response.
We need to study the basis filters in more detail. We are
interested in designing class I1 systems that operate on a
ki k2
relatively small neighborhood. This keeps the computationally
with the restrictions for 5 1 and k 2 as before for (44)1, and complexity low for both design and implementation. We
where we have substituted Okl,kz = - h z ( k l , k 2 , - k l , - 5 2 ) . choose a 5 x 5 region of support because it is a reasonable
Therefore, the set of filters in (44) completely describes all compromise between the computational complexity of the
possible filters from class 11, and it is not redundant since none filter and the degree of freedom for the design. We also have to
of the filters in (44) can be written in terms of any other from make sure that modeling the input signal by a single sinusoid

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 9% IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5 , NO. 6, JUNE 1996

remains a valid characterization and, thus, should keep the TABLE I

region of support as small as possible. We obtain a total of 12
basis filters for -2 5 kp 5 2 and 0 5 k1 5 2 in (44). We find k
- OT 0.1 0".
1.o
~

their frequency responses H i k 1 ' k 2 ) ( w2,

~ 1 ,ws, w g ) as 0 0.01323 1.01641
1 0.00061 0.04733 0.0
H i 0 q w l ,wp>wI(,w g ) = 1 - COS(W* - wg)
2 -0.00011 -0.00897 0.0
H ; o ' 2 ) ( W 1 , w2, W Q ,w g ) = 1 - cos(2w2 - 24) 3 0.00616 0.47368 0.5
H$l,-q)(wl, w.2;w 3 , w g ) = 1 - cos(w1 - 2w2 - w3 +2 4 ) 4 0.01325 1.01759 1.o
H;l 1 - 1)
( W l , wp, wg; w g ) = 1 - COS(W1 - w2 - wg + wg)
5 0.00616 0.47368 0.5
6 -0.0001 1 -0.00897 0.0
H p o ) ( w l ,w 2 , W Q ,w q ) = 1 - cos(w1 - w3)
7 -0.00021 -0.0 1674 0.0
H p ) ( w l , w z ; W j , W q )= 1 - cos(w1 + wp - w3 - wg) 8 -0.00012 -0.00961 0.0
H ; 1 , 2 ) ( w lw.2;
, w3, wg) = 1 - cos(w1 + 2w2 - w3 - 2 w 4 ) 9 0.00054 0.04197 0.0
10 -0.00012 -0.00961 0.0
H;2 1- 2)
( w 1 ,w.2;wg, w g ) = 1 - COS(2Wl - 2w2 - 2w3 + 2wg) 11 -0.00012 -0.01 674 0.0
- 1)( ~ 1~,p , w y , w g=) 1 - C O S ( ~ W-~ wp - 2 ~ +
3 wg)

H;2,")(w1:wp;w 3 , w q ) = 1 - cos(2w1 - 2w3)

We write the linear combination of the basis filters with
H ' p ) ( W I wp; WJ, wg) = 1 - cos(2w1 + wq - 2w3 - wq) coefficients Q k l , k a as a matrix product BO, where
H . p ( w 1 w.2 ~ w3, w g ) = 1- cos( 2w1 + 2wz - 2w3 - 2 4 .
We use the isotropy plots that we developed in Section
111-C as a characterization of class I1 filters. For each of
the basis filters, we obtain the response to a sinusoidal in- Thus, we can express the design problem as the linear matrix
put. For simplicity, we denote them B k l r k E( W O . U ) instead of equation
y;C,(kl,ka)
(WO , a). z=BH+e. (51)
Bo,1(wo, U) = 27r2(1 - cos(2wob))
The desired behavior of the ideal filter is contained in z,
B",Z(WO, U ) = 27r(1 - cos(4wob)) which we sample in the same fashion as described above for
B1,-2(wo, U ) = 2 7 4 1 - cos(2woa - 4wob)) B k l k z ( w ( ) .a ) , and e represents the error vector. We find the

131,-1(Wo, U ) = 27r(l - cos(2woa - 2wob)) optimal solution 8, which minimizes the error in the least-
squares sense [lS]
B1,o(wo, U) = 2 7 r ( 1 - cos(2w"a))
B1,1(wo, U) = 2n2(1 - COS(2W"U + 2wob)) H^ = [B'B] -'BTz. (52)
B1,2(wo, U) = 2 x 7 1 - cos(2woa + 4wob))
B~,-~(wO, U) = 27r2(1 - cos(4woa - 4wob)) We want to design a 2-D version of the Teager filter. In
B2,-1(wo, U ) = 2 7 4 1 - cos(4woa - 2wob))
[lo], it has been shown that Teager's algorithm responds
B~,o(w U)o=, 27rz(1 cos(4woa)) -
with an output y ( n ) = sin2(wo) if the input is a sinusoid
z ( n ) = sin(won). Now, we need to find a 2-D system that
B ~ , l ( w oa,) = 27r2(1 cos(4woa 2 w o b ) )
- + outputs a constant signal y(n1, na) = sin2(wo) for a sinusoidal
B ~ , ~ (Uw) = o ,2?(1 - c o s ( 4 w o ~ 4wob)) + input and is, of course, independent of the orientation of the
excitation.
where b = d m .Then, we sample these continuous We set z ( w o , u ) = sin2(wo) and convert z ( w o , a ) into the
functions in -1 5 U 5 1 and 0 5 W O 5 %. Since images
column vector z. Using (52), yields the parameter vector QT
tend to bc oversampled, very little information is contained
given in Table I. If we relax the restrictions somewhat and
in the very high frequencies, and for this design, we only
allow the filter output to be scaled, i.e., z(wg; U) = aisin2(wo),
consider frequencies up to 5.
We choose a sample spacing of
we can normalize the parameters such that E, 8 ( i ) = 3. This
Aa = 0.05 and Awo = 5,
which yields 201 . 26 samples.
yields 8, in Table I. Fig. 7(a) shows the isotropy plot for
We order the samples into column vectors in the following
manner. The first 201 elements belong to W O = 0 and from this filter. The design goal of virtually perfectly isotropical
behavior has been achieved up to W O = 5. If we had used the
U = -1 up to a = 1. After that, we take the 201 samples for
entire range from 0 to 7r for the optimization, then we would
W O = & and so on until we reach W O = and a = 1. We
have obtained a less accurate behavior for smaller frequencies.
number the functions B k l . k 2 ( W O , U ) in the order given before
The advantage of normalizing the parameter vector is that
from 1 to 12 and obtain vectors bl through b 1 2 , which we
combine into the matrix B the elements of 8, can easily be approximated by simple
numbers. The right-most column in Table I shows the approx-
B = (bl b p . .' b i z ) . (49) imation 4, using only the coefficients 1 and 0.5. Thus, we

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 959

Fig. 7. Ikotropy plots for the (a) 2-D Teager filter and (b) its approximation.

obtain the input-output relationship for the approximate 2-D the filter output is very small. Accordingly, undesired noise
Teager filter enhancement only takes place in the brighter portions, where
it is not as perceivable due to Weber’s law. The result of
enhancing the image with the unsharp masking algorithm in
Fig. I, and the quadratic Volterra filter in (53) as the edge
extraction filter is shown in Fig. 8(c). This image is noticeably
sharper, and even small details in the hair and feather regions
have been enhanced.
Another example is shown in Fig. 9. Again, the result
of unsharp masking with our Volterra filter in Fig. 9(c) is
noticeably sharper. Details have been extracted proportionally
to the local image brightness. In the darker portions of the
This equation is considerably simpler to implement than the
images (e.g., the tree region), the filter output in Fig. 9(bj is
full description in 6,. At the same time, the behavior of this
relatively small, whereas in the brighter portions, even minute
system as shown in Fig. 7(b) does not deviate too much from
details result in a strong response.
the ideal one. It is still extremely isotropic for almost the entire
In order to compare these results quantitatively with a
range of interest from 0 to 7r/2.
linear enhancement scheme using one of the commonly used
Laplacian filters, we need a criterion for image sharpness. We
I v . IMAGE PROCESSING APPLICATIONS have found that the mean magnitude of the gradient image
In this section, we want to demonstrate the application (using Sobel’s operator) corresponds well with the perceived
of quadratic filters to the processing of real images. The sharpness. In other words, if two images appear to be equally
designs in the previous sections were based on the idealized sharp, their mean gradient magnitude is approximately the
assumption that the input signal is a single sinusoid, which, of same.
course, is not true for a real image. The small region of support The measure that we use to assess the visibility of noise is a
of the filter, however, makes this approximation reasonable if relative one, that is, it needs a reference image for comparison.
we use the instantaneous frequency for the input sinusoid. The It observes the portion of the local grey-level fluctuation that
test image is a portion of “Lena” in Fig. 8(a) and (b) shows is already present in the reference (original) image and the
the output of the filter in (53). Details in the brighter parts part that is due to noise from the enhancement process. It also
of the image have been extracted, but in the darker areas, considers noise masking by the surrounding area of the image.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 960 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5 , NO. 6, JUNE 1996

(c)
Fig. 8. Original image “Lena” (a) and edge image obtained by applying the quadratic Volterra filter in (53) to the image in (a). Note that in darker
image areas, fewer edges are extracted than in the bright ones, which can be explained by the mean-weighted highpass characteristics of the filter. The
result of the unsharp-masking enhancement is shown in (c) with p = 0.002.

In highly active areas, the human visual system perceives by x and the enhanced one by y. Note that a comparison of
noise to a lesser degree than in flat ones. The local average the original image with itself yields k~ = 0. The value of k~
luminance influences the visibility as well. Altogether, we increases with the degree of image degradation and is (at least
combine these effects into a single number theoretically) not bounded.
Since the original “Lena” and “Park” images are very
(54) clean, white Gaussian noise ( p n = 0.0, 02 = 100.0) was
added to both of them. We then enhanced these images
The index 1 stands for local and indicates that the variances using the Volterra filter in (53), the Volterra filter C in (40)
and the mean are dependent on n1 and 712 and must be (which is called Type 1A in [ l l ] ) , and the Laplacian filters,
computed over a local neighborhood (5 x 5 pixels for our such that the resulting overall image sharpness is the same
simulations). The sum is carried out over all pixels in the in all corresponding cases. Table I1 lists the quantitative
image. Even though the human eye cannot adapt to very small comparisons. Adding noise to an image does not increase
portions of the image, we assume that the observer would view pllg1l significantly, which corresponds well with our visual
the image at arbitrary distances, which justifies the use of a observations. We then computed the noise visibility measure
localized measure. The reference (original) image is denoted k,v relative to the noisy image before enhancement. The

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 ~

THURNHOFER AND MITRA GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 96 1

(c)
Fig. 9. (a) Original image “Park” and (b) edge image obtained by applying the quadratic Volterra filter in (53) to the image in (a). The result of the
unsharp-masking enhancement is shown in (c) with p = 0.002.

Volterra filter C shows only slightly increased noisyness since TABLE 11

it also belongs to class of mean-weighted highpass filters.
The Laplacian filters, however, lead to values of k~ that are
considerably larger than for the Volterra filter in (53), which
“Lena”
corresponds to a significant increase in noisyness. To show
an example of this problem, Fig. 10 compares the results
for the “Park” image visually. The result using the Volterra
filter C is similar to the image in (d). Even though the three
enhanced images have the same overall sharpness, we can see
that Figs. 10(b) and (c) are more noisy than (d), which is most “Park”
noticeable in the dark portions of the trees.

V. CONCLUSION
We have developed a subclass of 1-D quadratic Volterra
filters that can be written as a product of local mean and linear found before heuristically, we are now able to develop them
highpass. Even though some filters of this class have been in a unified framework. We have also extended this concept

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 962 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 5, NO. 6, JUNE 1996

(c) (d)
Fig. 10. Comparison of noise enhancement using u n s h q masking. The “Park” image with added Gaussian noise is shown in (a). The results of unsharp
masking using the first Laplacian filter in (b) ( p = 0.45),the second Laplacian filter in (c) ( p = 0.17),and the Volterra filter in (53) in (d) ( p = 0.002)
demonstrate that linear filters amplify noise, especially in the dark areas more than our Volterra filters. All the enhanced images have approximately the
same sharpness. The Laplacian I11 filter performs very poorly, and the resulting image in not shown here.

to 2-D systems obtaining the general class of filters for edge quantitative criteria, we have shown that quadratic Volterra
extraction. Describing these systems using the kemel and the filters yield perceptually better results than standard linear
frequency response alone, however, does not enable us to find filters.
the coefficients in a simple and robust way. Thus, we have
developed an approximate description that is more suitable
for the problem at hand by using simple 2-D inputs. This REFERENCES
also allows us to evaluate the orientational dependence of the [ 11 K. R. Boff, L. Kaufman, and J. P. Thomas, Handbook of Perception and
filter. We have designed an optimal filter using basis filters and Human Performance, vol. I: Sensory Processes and Perception. New
York: Wiley, 1986.
a least-squares frequency domain approach. Compared with [2] J. S. Lim, 2 - 0 Signal and Image Processing. Englewood Cliffs, NJ:
other methods [4], [ 6 ] , [9], [16], [17], this technique gives a Prentice-Hall, 1990.
131 A. K. Jain, Fundamentals of Digital Image Processing, Englewood
more intuitive understanding of the filter behavior and does Cliffs, NJ: Prentice-Hall, 1989.
not depend on heuristic assumption for the impulse response. [4] G. L. Sicuranza, “Quadratic filters for signal processing,” Proc. IEEE,
vol. 80, pp. 1263-1285, Aug. 1992.
As an immediate application in image processing, we have [5] I. Pitas and A. N. Venetsanopoulos, Nonlinear Digital Filters. Boston,
considered the unsharp masking enhancement scheme. Using MA: Kluwer, 1990.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.
 THURNHOFER AND MITRA: GENERAL FRAMEWORK FOR QUADRATIC VOLTERRA FILTERS FOR EDGE ENHANCEMENT 963

161 S. A. Billings and K. M. Tsang, “Spectral analysis for nonlinear Sanjit K. Mitra (S’59-M’63-SM’69-F‘74) re-
systems, Part I: Parametric nonlinear spectral analysis,” Mech. Syst. ceived the B.Sc. degree (Hons.) degree in physics
Signal Processing, vol. 3, no. 4, pp. 319-339, 1989. in 1953 from Uktal University, Cuttack, India,
[7] L. 0. Chua and C. Y. Ng, “Frequency domain analysis of nonlinear the M.Sc. (Tech.) degree in radio physics and
systems: general theory,” Electron. Circuits Syst., vol. 3, pp. 165-185, electronics in 1956 from Calcutta University,
July 1979. Calcutta, India, and the M.S. and Ph.D. degrees
[XI M. Schetzen, The Volterra and Wiener Theories of Nonlinear Systems. in electrical engineering from the University
New York: Wiley, 1980. of Califomia, Berkeley, in 1960 and 1962,
[9] G. F. Ramponi, G. L. Sicuranza, and W. Ukovich, “A computational respectively.
method for the design of 2-D Volterra filters,” IEEE Trans. Circuits From June 1962 to June 1965 he was with Comell
Syst., vol. 35, pp. 1095-1102, Sept. 1988. University, Ithaca, NY, as a Professor of Electrical
[IO] J. F. Kaiser, “On a simple algorithm to calculate the ‘energy’ of a Engineering. ]He was with AT&T Bell Laboratories, Holmdel, NJ, from June
signal,” in Proc. IEEE Int. Con$ Acoust., Speech Signal Processing, 1965 to Janu.my 1967. He has been with the faculty of the University of
Albuquerque, NM, Apr. 1990, pp. 381-384. California since then, first at the Davis campus and more recently at the Santa
[ l I] S. K. Mitra, H. Li, I. S. Lin, and T.-H. Yu, “A new class of nonlinear Barbara campus as a Professor of Electrical and Computer Engineering,
filters for image enhancement,” in Proc. IEEE Int. Conf: Acoust, Speech where he served as Chairman of the Department from July 1979 to June
Signal Processing, Toronto, Ont., Canada, 1991, pp. 252552528, 1982.
1121 S . Thumhofer, “Quadratic Volterra filters for edge enhancement and Dr. Mitra served as President of the IEEE Circuits and Systems
their applications in image processing,” Ph.D. dissertation, Univ. of Society in 1986. He is currently a member of the editorial boards of the
Califomia, Santa Barbara, 1995. Intemational Joumal on Circuits and Systems and Siganl Pvocessing, the
[I31 S. Thumhofer and S. K. Mitra, “Volterra filters for perceptual edge Intemational Joumal on Multidimensional Systems and Signal Processing,
extraction,” in Proc. 28th Ann. Asilomar Confi Signals, Syst. Comput., Signal Processing, and the Joumal of the Franklin Institute. He received the
Pacific Grove, CA, Nov. 1994, pp. 736-740. 1973 F. E. Terman Award and the 1985 AT&T Foundation Award of the
[14] S. Thurnhofer and S. K. Mitra, “Designing quadratic Volterra filters American Society of Engineering Education, the Education Award of the
for nonlinear edge enhancement,” in Proc. Inl. Con$ Digital Signal
IEEE Circuits and Systems Society in 1989, and the Distinguished Senior
Processing, Limassol, Cyprus, June 1995, pp. 320-325.
U.S. Scientist Award from the Alexander von Humboldt Foundation of West
[ 151 J. M. Mendel, Lessons in Digital Estimation Theory. Englewood Cliffs,
Germany in 1989. In May 1987, he received an Honorary Doctorate of
NJ: Prentice-Hall, 1987.
[ I 61 G. Ramponi, “Bi-impulse response design of isotropic quadratic filters,” Technology degree from Tampere University, Tampere, Finland. He is a
Proc. IEEE, vol. 78, pp. 665-677, Apr. 1990. Fellow of the AAAS and SPIE and is a member of EURASIP and ASEE.
[I71 G. Ramponi, G. L. Sicuranza, and W. Ukovich, “An optimization
approach to the design of nonlinear Volterra filters,” in Proc. EUSIPCO-
86, Signal Processing III: Theories Applications,, The Hague, The
Netherlands, Sept. 1986, pp. 151-154.

Stefan Thurnhofer (M’96) received the Dip1 -1ng

degree in electncal engineering from the University
of Erlangen-Nurnberg, Germany, in 1991 and the
Ph.D. degree in electrical and computer engineering
in 1995 from the University of Califomia, Santa
Barbara (UCSB).
His research interests at UCSB were in digital
signal and image processing, the design and appli-
cation of nonlinear enhancement methods for digital
images using Volterra filters, image interpolation,
and error diffusion halftoniug In 1995, he joined
Lucent Technologies (formerly, AT&T Bell Laboratone\), Allentown, PA,
where he is currently involved in the design of microcontrollers and digital
signal processors

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KHARAGPUR. Downloaded on March 8, 2009 at 22:01 from IEEE Xplore. Restrictions apply.