Precision Engineering

Journal of the International Societies for Precision Engineering and Nanotechnology

26 (2002) 222235

Recent advances in separation of roughness, waviness and form

J. Raja , B. Muralikrishnan, Shengyu Fu
Center for Precision Metrology, University of North Carolina at Charlotte, 9201 University City Blvd, Charlotte, NC 28223, USA
Received 17 September 2001; accepted 6 December 2001

Abstract
Engineering surfaces are comprised of a range of spatial wavelengths. Filtering techniques are commonly adopted to separate the
different wavelength components into well-defined bandwidths. Filtering is done prior to numerical characterization and it is also essential
for extracting information needed to provide process feedback and establish functional correlation. This paper reviews commonly used
filters in surface metrology like the 2RC, Gaussian and several new ones currently under research such as the spline, morphological,
wavelets, regression filters and robust regression filters. The need for these new filters and examples illustrating the features of these filters
are also presented. 2002 Elsevier Science Inc. All rights reserved.
Keywords: Surface metrology; Filters; Roughness; Waviness

1. Introduction
A typical engineering surface consists of a range of spatial frequencies. The high frequency or short wavelength
components are referred to as roughness, the medium
frequencies as waviness and low frequency components
as form. Fig. 1 illustrates this. In many cases the original profile may also contain significant form such as a
radius. In the early days, different instruments and measurement techniques were adopted to capture the different
wavelength regimes. Visual assessment using microscopy
was used to identify any torn material or micro burr.
Stylus-based instruments were used to obtain roughness.
Other special instruments obtained form information. Today, with the availability of increasingly sophisticated
gages, sensors and powerful data processing capabilities
using personal computers, there is an overlap of measuring capability. A typical stylus-based instrument can
capture roughness, waviness and form. A roundness measuring instrument can also gather straightness data and
a CMM can get both dimensional information and form.
Fig. 2 shows the current metrology measurement spectrum. As the bandwidth of measurement instruments increases, it becomes essential to separate surface profile
data into meaningful wavelength regimes before numerical
characterization.

Corresponding author.
E-mail address: jraja@uncc.edu (J. Raja).

Historically, it has been accepted that different aspects of

the manufacturing process generate the different wavelength
regimes and these affect the function of the part differently
[3]. However, there are no case studies that conclusively
prove this relationship. By separating surface profile into
various bands, it is possible to map the frequency spectrum
of each band to the manufacturing process that generated
it. Thus, filtering of surface profiles serves as a useful tool
for process control and diagnostics. While engineers commonly trace manufacturing process variations based on surface profile data, mapping the functional performance of a
component based on surface profile information has been a
challenge. The different wavelength regimes play a key role
in critical parts like crankshafts and bearings. Thus, separation of signal into various bandwidths has to be viewed from
a functional standpoint as well. Recent trend in outsourcing of manufacturing makes it essential to define surface
texture requirements very clearly. This has made it necessary to use filtering techniques to establish the required
wavelength regimes before numerical characterization. In
addition, meaningful comparison of parameters from different surface texture measuring instruments can only be done
if filtering techniques are used to establish identical bandwidths. Hence filters used in surface texture instruments have
become critical for commerce, process diagnostics and to
establish functional correlation. This paper reviews the evolution of current filters that are defined in the standards and
the need for new filters. The features of the new filters are
presented through examples.

0141-6359/02/$ see front matter 2002 Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 1 - 6 3 5 9 ( 0 2 ) 0 0 1 0 3 - 4

J. Raja et al. / Precision Engineering 26 (2002) 222235

223

Fig. 1. Roughness and waviness in a surface [1].

Fig. 2. Metrology measurement spectrum [2].

2. Filters currently used to characterize surface

profiles
2.1. 2RC filter
The earliest filter used in surface metrology is the analog
2RC filter. The highpass 2RC filter can be realized digitally
using the weighing function given in Eq. (1)






A
|x|
|x|
S(x) =
2A
exp A
(1)
c
c
c
where A = 3.64 for 75% transmission at the cutoff, x the
position from the origin of the weighing function ( <
x < 0) and c is the long wavelength roughness cutoff.
The mean line is obtained by convolving the profile with
the weighing function in Eq. (1). The mean line is then
subtracted from the original profile to obtain the roughness
profile. The transfer function of the 2RC highpass filter is



output
2
(2)
= 1 ik
input
c

where i = 1 and k = 1/ 3 = 0.577. Fig. 3 shows the

weighing function and the transmission characteristics of a
2RC highpass filter.
The major disadvantage of the 2RC filter is its nonlinear
phase. An example illustrating the phase distortion effect of
the 2RC filter is shown in Fig. 4. The mean line (0.8 mm
cutoff) and the profile are shown in Fig. 4(a). The roughness
profile is shown in Fig. 4(b). From Fig. 4(a), it can be seen
that the mean line has a phase offset which causes considerable distortion in the roughness profile. This effect becomes
more severe as the cutoff increases.
To overcome the problem of phase distortion, Whitehouse
[4] introduced a phase-correct filter. While zero phase is
unattainable, a filter with symmetrical weighing function can
be designed to yield linear phase. Raja and Radhakrishnan
[5] have explored the use of digital filters, particularly finite
impulse response filters (FIR) that have linear phase for
surface profile filtering.
Another disadvantage of the 2RC filter is that waviness
cannot be obtained by simply subtracting the profile from
roughness because the transmission at the cutoff is 75%.

224

J. Raja et al. / Precision Engineering 26 (2002) 222235

Fig. 3. Weighing function and transmission characteristic of 2RC highpass filter.

Therefore, a complementary waviness filter that transmits

75% at the cutoff is necessary to capture waviness. It is
a normal practice to consider the mean line obtained by
convolving a profile with the weighing function given in
Eq. (1) as the waviness. However, true waviness is obtained
by convolving the profile with the weighing function given
in Eq. (3)
s(x) =

xA2 xA/c
e
,
2c

0 < x < ,

eliminated this filter from their standards. A new filter, referred to, as the Gaussian filter was developed to overcome
two primary drawbacks of the 2RC filter, nonlinear phase
and the need for simpler way to implement both a roughness and a waviness filter. The characteristics of this filter
are discussed in Section 2.2.
2.2. Gaussian filter

A = 2 3
(3)

where x is the position from the origin of the weighing

function and c is the long wavelength roughness cutoff.
The definition of a digital filter provides a direct implementation for 2RC filter using convolution [1]. While it is
clear, simple and versatile, more storage is needed and the
computational speed is low for this method. An alternative
recursive technique is discussed by Raja and Radhakrishnan
[5] and by Whitehouse [6] to overcome this problem. While
the ASME B46.1 [1] still includes the 2RC filter, ISO has

The most widely used filter today is the Gaussian filter,

which is described in ASME B46.1 [1] and ISO 11562 [7].
The weighing function and transmission characteristic of a
Gaussian lowpass filter are given by



 
1
x 2
S(x) =
exp
(4)
c
c



 
Aoutput
c 2
= exp
,
Ainput

Fig. 4. (a) Profile and 2RC mean line (not waviness profile); (b) roughness profile.

(5)

J. Raja et al. / Precision Engineering 26 (2002) 222235

225

Fig. 5. Weighing function and transmission characteristic of Gaussian filter.

respectively, where = ln 2/ = 0.4697, x is the position

from the origin of the weighing function, c is the long
wavelength roughness cutoff and is the wavelength of
different sinusoidal profiles on the surface. They are shown
in Fig. 5.
An important property of Gaussian filter is its linear phase,
which is a major advantage over 2RC filter. Also, the filter
is designed to have 50% transmission at the cutoff. This allows the calculation of waviness by simply subtracting the
roughness from the raw profile, another important improvement over the 2RC filter. An example profile and the Gaussian mean line (0.8 mm cutoff) are shown in Fig. 6. The 2RC
mean line is also shown to illustrate the linear phase of the
Gaussian filter and the nonlinear phase of the 2RC filter.
The boundary distortion due to the local weighed average
is inevitable for both 2RC and Gaussian filters. Therefore,
the mean line at the boundary region cannot be used for
evaluation purposes. Generally, the first half and the last half
cutoff length are discarded to eliminate end effects. Convolution is the most common approach for implementing Gaussian filter [1,6]. For increasing the speed, implementation
using FFT was introduced by Raja and Radhakrishnan [8].
Another fast and reliable convolution algorithm for Gaussian
filter has been derived by Krystek [9] based on a recurrence
relation for the weighing function using the symmetry condition. Other algorithms have also been investigated by many
researchers. For instance, a new fast algorithm for construct-

ing a series of Gaussian filters according to the approximation of Gaussian function and central limit theorem was applied by Yuan et al. [10]. The objective of this approach is
to reduce the multiplication operations in the procedure to
increase calculation efficiency.
While the Gaussian filter is an improvement over the 2RC,
there are still several issues with the use of this filter. Edge
effects prevent the use of the first and last cutoff. This filter
also performs poorly on profiles with deep valleys such as
the ones found in plateau honed surfaces. An empirical filtering method referred to as Rk filter was developed to deal
specifically with plateau honed surface profiles.
2.3. Rk filter
The Gaussian filter is not robust against outliers. Special
applications such as plateau honing require a filter whose
mean line is not distorted by the valleys. ISO 13565 [11]
recommends a two-step filtering approach using a Gaussian
phase correct filter in accordance with ISO 11562. In the
first step, the primary profile is filtered using a low pass
Gaussian filter. All points in the primary profile that lie below
the mean line are replaced by the mean line itself. Thus,
all valleys are suppressed for the second step. The modified
profile is sent through the same Gaussian low pass filter and
the new mean line obtained becomes the final mean line.
The primary profile is subtracted from this mean line to

Fig. 6. Gaussian mean line.

226

J. Raja et al. / Precision Engineering 26 (2002) 222235

Fig. 7. Rk and Gaussian mean line on a profile with deep valleys.

obtain the roughness profile. A profile with deep valleys is

shown in Fig. 7 with the mean line generated by Gaussian
and Rk filter with a 0.8 mm cutoff. The Gaussian mean line
is pulled down near the valleys while the Rk mean line hugs
the texture. The effect of the Gaussian mean line on the
roughness profile is shown in Fig. 8. Artificial features are
introduced into the roughness profile because the filter is not
robust against outliers/valleys.
DIN standards [12] recommend a triangular weighing
function as shown in Fig. 9. The roughness amplitude transfer function is given by


2



Aoutput

c
=1
sin2
(6)
Ainput
c

where is the wavelength of the sine wave to be filtered,

= 0.44294647 and c = B/2. The two-step filtering
procedure is the same as that described in ISO 13565.

Rk filter suffers from the same edge effect problems mentioned before. Also the two-step iteration process is not
always robust against very deep valleys. Splines, robust
splines, regression and robust regression filters offer solutions to these problems. Subsequent sections present them
in greater detail.

3. Advanced filters for surface profile filtering

3.1. Spline filter
Spline filter was introduced to overcome the unrealistic
separation of roughness in surfaces with large form and
the end effects of both 2RC and Gaussian filter. There
are two kinds of spline filters: non-periodic and periodic.
Non-periodic splines are used for filtering open profiles and

Fig. 8. Roughness profile distortion effect.

Fig. 9. Transfer characteristics and weighing function.

J. Raja et al. / Precision Engineering 26 (2002) 222235

227

Fig. 10. Lowpass transmission characteristic (x = c /200).

periodic splines for closed profiles. The weighing function

of a spline filter cannot be given by a simple closed formula.
Therefore, filter equations are used instead of weighing
functions to describe spline filters. The filter equation for
the non-periodic spline filter is given by
(1 + 4 Q) = z

(7)

where

Q=

4
1

5
2

2
1 nn

1
a1
4
4 x
= 1 + 16 sin
a0

(10)

Fig. 10 shows the transmission characteristics and Fig. 11

shows a mean line obtained from a spline filter applied to
a profile with large form. The absence of edge effects in
the spline is notable. Spline filters are described in detail
by Krystek [13,14]. ISO/TC 213 is currently working on a
draft document to include spline filters into their standards.
3.2. Robust spline filter

(8)

and is given by




1
x
= 2sin
c

The lowpass transmission characteristics is given by

(9)

Robust spline filter is a modification of the spline filter to

overcome the problem of large form and end effects while
being robust against deep valleys. This filter is currently
being discussed in the ISO/TC 213 committee.
The spline filter has taken care of two drawbacks of the
Gaussian filterthe problem of large form and end effects.
Recent research work has provided another solution to this
problem by modifying the implementation of the Gaussian
filter, namely Gaussian regression filter. The Gaussian regression filter is discussed in Section 3.3.

Fig. 11. Comparison between Gaussian and spline filter.

228

J. Raja et al. / Precision Engineering 26 (2002) 222235

3.3. Gaussian regression filter

The Gaussian regression filter [15,16] is a modification
of the Gaussian filter in order to evaluate the entire length
of the profile. The Gaussian regression filter is defined by
the following general regression arrangement:
n


(zl wk )2 skl x Min

wk

(11)

where z is the profile ordinate; w the mean line ordinate;

n the number of points; k the index for the location of the
weighing function; l the index for profile points and s is the
weighing function as given by



2 ((k l) x)2
1
skl =
exp
(12)
ln(2)
2c
c ln(2)
c is the cutoff.
In the regression approach, the weighing function is calculated for every point on the profile and the objective function
is minimized to compute the ordinate on the mean line w. A
zero-order regression filter minimizes the square of the deviation of the profile ordinate from a horizontal line at each
point. Thus, the minimization problem involves computing a
constant term at every point of the profile. The second-order
regression filter minimizes the square of the deviation of the
profile ordinate from a quadratic curve at each point. Here,
the minimization involves finding three coefficients of the
quadratic curve at every point of the profile.
3.4. Zero-order Gaussian regression
The zero-order regression filter can be mathematically
shown to be equivalent to the convolution given in Eq. (13)
w(l) =

z(i)sreg (l i)

as shown in Fig. 12. However, it remains unchanged in the

interior regions of the profile. Fig. 13 shows a profile with
a Gaussian and a zero-order Gaussian regression mean line.
Notice the absence of end effects in the regression filter. The
regression filter is described in detail by Brinkmann et al.
[15,16].
3.5. Second-order Gaussian regression
The second-order Gaussian regression filter [16] minimizes the deviations between the input dataset and a
quadratic term to obtain the ordinate of the mean line
at each point of the profile. The second-order filter, like
the spline, can be used for form removal. However, it is
computationally expensive, as the regression process requires polynomial fitting for every point. Fig. 14 shows a
Gaussian and second-order Gaussian regression mean line
on a curved profile.
3.6. Robust Gaussian regression filter

(13)

i=1

s(l i)
sreg (l i) = 
s(l i)

Fig. 12. Progression of weighing function.

(14)

From Eq. (14), it can be seen that the weighing function

of this filter is modified in the first and last cutoff sections

The robust algorithm [15] applies a regression filter iteratively to a dataset until the mean line is satisfactory. A
median statistic is used as the index to decide when to stop
the iteration. The filter can be applied to datasets with outliers that would distort a simple Gaussian mean line. While
the Rk technique applies the Gaussian window twice to the

Fig. 13. Gaussian and zero-order Gaussian regression mean line, cutoff = 0.8 mm.

J. Raja et al. / Precision Engineering 26 (2002) 222235

229

Fig. 14. Gaussian and second-order Gaussian regression mean line, cutoff = 4 mm.

Fig. 15. Robust zero-order Gaussian regression and Rk mean line.

dataset, the robust Gaussian regression algorithm applies a

regression filter several times depending on the end condition criteria. The robust algorithm is given by
n

1

(zl wk )2 skl i,l x Min

wk

(15)

where i,l is an additional vertical weight, i the iteration

number and l is the ordinate index of the profile. In the first
iteration, i,l is set to unity. In subsequent iterations, i,l is
computed as follows:



2 2
ri,l
ri,l
i+1,l = 1
, for < 1
(16)
CB
CB
otherwise
i+1,l = 0,
where CB = 4.4 median(zl wl ). Fig. 15 shows the mean
line generated by the robust algorithm and the Rk procedure.
The regression approach has resulted in an improved mean
line at the end.
ISO/TC 213 is currently working on a draft document to
incorporate regression and robust regression filters into their
standards. To summarize, regression filters permit evaluation of the entire length and higher order regression filters
perform better than Gaussian filters on profiles with large
form. Robust filters provide better mean line for profiles
with deep valleys.
Having discussed several filters and their characteristics,
it is worthwhile to note that traditionally, a single cutoff
value has been used to separate roughness from waviness
and another cutoff to separate waviness from form. However, with increasing overlap of measurement capabilities
of current instruments and the emerging trend of using sur-

face profiles for process diagnostics, there is a need for the

separation of signal into more meaningful bandwidths. This
has resulted in the development of multi-band filters and filter banks. One recent development in this regard is the use
of wavelets for surface profile analysis. Wavelet analysis is
discussed in Section 4.

4. Wavelet analysis and wavelet-based filters

Partitioning a signal into multiple bands for purposes
of providing process feedback or functional correlation
can become a computationally expensive task. Also, commonly used filters such as the Gaussian filter are not
suitable for such applications because of their poor transmission characteristics. Wavelets provide good separation
of signal into multiple bands in a computationally efficient
manner.
Wavelets are mathematical functions that partition data
into different frequency components, enabling us to study
each component with a resolution matched to its scale. They
have advantages over traditional Fourier methods in analyzing physical situations where the signal contains discontinuities and sharp spikes. The fundamental idea behind wavelet
analysis is to analyze signal according to scale. Wavelet
analysis provides a flexible time-scale window localized on
time-scale plane. Gross features would appear when looking at a signal with a large window. Similarly, small features would appear when the signal is viewed with a small
window. Wavelets permit looking at both the small and the
large features simultaneously. This makes wavelets interesting and useful.

230

J. Raja et al. / Precision Engineering 26 (2002) 222235

Fig. 16. Multi-resolution analysis for surface profiles [20].

Chen et al. [17] and Liu and Raja [18] presented an early
study on the application of multi-resolution analysis (MRA)
for analyzing multi-scale engineering surfaces. As octave
filter banks, wavelets can be used to decompose signal into
different bands with different scales, which is also called
multi-resolution analysis of a signal [19]. In MRA, the dyadically dilated wavelets constitute a bank of octave band pass
filters and the dilated scaling functions form a low pass filter bank. The efficient implementation of discrete wavelet
transform (DWT) is to get successive approximations of a
signal by applying the low pass filter bank and successive
details of a signal by applying the band pass filter bank.
DWT appears to have a great potential for analyzing the
multi-scale features in engineering surfaces due to its properties of good time-scale localization and flexible time-scale
resolution. In order to analyze engineering surface texture,
multi-scale approximations of original texture at different

Fig. 17. MRA approximations and details of a piston pin bore surface profile.

Fig. 18. MRA reconstruction at levels: 15, 59, 913.

J. Raja et al. / Precision Engineering 26 (2002) 222235

231

Fig. 19. Lifting stages: split, predict and update.

resolution levels will be extracted by DWT. This will give a

clear overview of the multi-scale features of surface texture.
Then the separation of multi-scale features based on wavelength can be made according to the information provide by
the multi-scale approximations. Fig. 16 shows the general
procedure to do MRA for a surface profile.
Due to the linear phase requirement of surface analysis,
biorthogonal wavelet family is usually applied for wavelet
analysis. Due to the scope of the paper, the mathematical
theory of design and implementation of the biorthogonal
wavelets is referred to Vetterli and Herley [21]. Figs. 17
and 18 show an MRA analysis using biorthogonal wavelets
using the wavelet toolbox in Matlab.
In 1994, the lifting scheme was developed by Sweldens
[22], which leads to so-called Second Generation
Wavelets. Lifting leads to algorithms that can easily be
generalized to complex geometric situations, which typically occurs in engineering surface analysis. The lifting
scheme is a new approach for the construction of biorthogonal wavelets entirely in the spatial domain. This is in
contrast to the traditional approach, which relies heavily
on the frequency domain. The lifting scheme consists of
three stages: split, predict and update, which are shown in
Fig. 19. An application of the lifting scheme for surface
profile analysis is presented by Jiang et al. [23].
Wavelet analysis is a logical extension of a simple filter. It permits us to analyze signal according to scale. In

Section 5, an entirely different class of filters is discussed

Morphological filter, which is a superclass of the envelope
filters of the late-1970s. While these filters provide envelope
mean lines and are only affected by the peaks or the valleys, they can also be used to build a ladder structure that
makes them very similar to wavelets. Morphological filters
and scale-space techniques are discussed next.
5. Morphological filters
A discrete morphological filter [24] takes two inputs, the
profile z and the structuring element s and produces a filtered profile w. Dilation and erosion are two fundamental
operations. Dilation expands the input set by the structuring element. Erosion shrinks the input set by the structuring
element. A typical surface profile recorded by any stylus instrument is the result of dilation of the true surface by the
stylus tip. Closing and opening are secondary operations.
Closing is obtained by dilation followed by erosion. Opening is obtained by erosion followed by dilation. Closing filter
produces an envelope mean line that could be useful in applications involving rolling ball/sliding plane contact simulations. Fig. 20 shows a profile and mean lines generated by
closing filters, one with a straight-line structuring element
of size of 0.1 mm and another with 0.05 mm. It is obvious
that the mean line generated using a larger line will ride on

Fig. 20. Morphological closing filter mean lines.

232

J. Raja et al. / Precision Engineering 26 (2002) 222235

Fig. 21. Alternating symmetrical filter representation.

the larger peaks and is not influenced by valleys whose size

is smaller than its length.
Morphological filters lend themselves to outlier analysis. A closing filter of a given scale will remove all valleys

whose widths are smaller than this scale. An opening filter will remove all peaks whose widths are smaller than the
scale of the filter. An alternating symmetrical filter [25] is a
combination of a closing and opening filter and can be used

Fig. 22. Difference profiles.

Fig. 23. Approximation profiles.

J. Raja et al. / Precision Engineering 26 (2002) 222235

233

Fig. 24. Process analysis example.

to remove both peaks and valleys that are smaller then the
scale of the filter. A series of alternating sequence filters can
be applied to a profile, starting with a scale that is slightly
larger than the stylus tip. As the scale is graduallyincreased,
the successively smoother profiles can be plotted on a ladder structure. This ladder structure is useful in visualizing

surface creation and also helps in identifying and removing

outliers.
The structure of an alternating symmetrical filter is
very similar to multi-resolution analysis in that the profile
is successively viewed at different scales. At each level,
an approximation and a difference profile is generated.

Fig. 25. Functional correlation example.

234

J. Raja et al. / Precision Engineering 26 (2002) 222235

The approximation is then analyzed at a larger scale. The

schematic of an alternating symmetrical filter is shown in
Fig. 21. Fig. 22 shows an example profile and the difference profiles generated by using ball structuring elements
of different sizes. Fig. 23 shows the approximation profiles at each level. The final profile can be reconstructed
by simply adding all the differences and the approximation of the last level. Partitioning of the signal into various scales and reconstruction is very similar to wavelet
analysis.
ISO/TC 213 is currently drafting a report on morphological filters and scale-space techniques for their standards.
The preceding sections presented several filters and their
characteristics. After having looked at the different filters,
one question remains to be answered. Why filter surface
profiles? The answer is tied to the fundamental reason for
measuring surface profilesfor functional correlation and
process diagnostics. The last section presents an example
illustrating process diagnostics and functional correlation in
surface metrology.

6. Application of filtering in process diagnostics and

functional correlation
To analyze surface texture appropriately, the different
wavelength components should be separated reasonably by
applying filters with different cutoffs. Fig. 24 is a piston
pin bore profile and its FFT plot from a study by Malburg
[2]. The FFT plot shows the different wavelengths and their
sources. The short wavelengths are generated by tool radius
and long wavelengths by slideway error. Chatter is clearly
visible in the medium wavelength band.
Functional performance of a part is usually determined
by different wavelengths on a surface. From a study by Malburg [2], the middle wavelength components are critical to
a diesel engine crankshaft pins performance because they
are relative to localized loading and ultimate failure of the
bearing. It is necessary to control long wavelength aspects in
a reasonable level for proper fit and short wavelength components for tribological requirements. Bandpass filtering is
needed to separate those function critical wavelength components. Fig. 25 shows the result from bandpass filtering of
a crankshaft pin profile.

7. Conclusions
Advances in filtering techniques enable us today to
separate the different wavelength regimes without any distortions. It is easy to specify and implement well-defined
bandwidths using digital filters. Current research emphasis is in the area of multi-resolution analysis and more
robust filters. The need to establish stable manufacturing
process and provide functional correlation for components
will require finer bandwidth analysis in the future. The

new filters provide a powerful toolbox to move surface

texture analysis beyond simple parameter calculation based
on one cutoff. Number of groups in the world are using
these techniques that could lead to case studies demonstrating manufacturing process control and functional
correlation.
References
[1] ASME B46.1. Surface texture (surface roughness, waviness, and lay),
1995.
[2] Malburg MC. A unified methodology for the application of surface
metrology. Ph.D. dissertation, University of Warwick, UK, and
University of North Carolina at Charlotte, USA, 1996.
[3] Whitehouse DJ. Surfacesa link between manufacture and function.
Proc Inst Mech Eng 1978;192:17988.
[4] Whitehouse DJ. Improved type of wavefilter for use in surface-finish
measurement. Proc Inst Mech Eng 196768;182:30618.
[5] Raja J, Radhakrishnan V. Digital filtering of surface profiles. Wear
1979;57:14755.
[6] Whitehouse DJ. Handbook of surface metrology. Bristol, UK:
Institute of Physics Publishing, 1994.
[7] ISO 11562. Geometrical product specification (GPS)surface
texture: profile methodmetrological characteristics of phase correct
filters. Geneva: International Organization for Standardization, 1996.
[8] Raja J, Radhakrishnan V. Filtering of surface profiles using fast
Fourier transform. Int J Mach Tool Des Res 1979;19:13341.
[9] Krystek M. A fast Gauss filtering algorithm for roughness
measurements. Prec Eng 1996;19:198200.
[10] Yuan YB, Qiang XF, Song JF, Vorburger TV. A fast algorithm for
determining the Gaussian filtered mean line in surface metrology.
Prec Eng 2000;24:629.
[11] ISO 13565-1. Geometrical product specification (GPS)surface
texture: profile method surfaces having stratified functional
properties. Part 1. Filtering and general measurement conditions,
1996(E).
[12] DIN 4776. Measurement of surface roughnessparameters Rk , Rpk ,
Rvk , Mrl , Mr2 for describing the material portion in the roughness
profile, 1990.
[13] Krystek M. Discrete linear profile filters. In: Proceedings of the
Xth International Colloquium on Surfaces. Chemnitz University of
Technology, Chemnitz, 2000. p. 14552.
[14] Krystek M. Form filtering by splines. Measurement 1996;18:915.
[15] Brinkmann S, Bodschwinna H, Lemke HW. Development of a robust
Gaussian regression filter for three-dimensional surface analysis.
In: Proceedings of the Xth International Colloquium on Surfaces.
Chemnitz University of Technology, Chemnitz, 2000. p. 12232.
[16] Brinkmann S, Bodschwinna H, Lemke HW. Accessing roughness in
three-dimensions using Gaussian regression filtering. In: Proceedings
of the 8th International Conference on Metrology and Properties of
Engineering Surfaces, in Association with IMEKO. University of
Huddersfield, UK, 2000.
[17] Chen X, Raja J, Simanapalli S. Multi-scale analysis of engineering
surface. Int J Mach Tools Manufact 1995;35(2):2318.
[18] Liu X, Raja J. Analyzing engineering surface using wavelet filter.
In: Proceedings of the SPIE. Vol. 2825, 1996.
[19] Mallat S. A theory for multi-resolution signal decomposition:
the wavelet representation. IEEE Trans Pattern Anal Mach Intell
1989;11(7):67493.
[20] Liu X. Advanced techniques for separation of roughness, waviness
and form. Ph.D. dissertation, University of North Carolina at
Charlotte, 1999.
[21] Vetterli M, Herley C. Wavelets and filter banks: theory and design.
IEEE Trans Signal Proc 1992;40(9):220732.

J. Raja et al. / Precision Engineering 26 (2002) 222235

[22] Sweldens W. The lifting scheme: a custom-design construction of
biorthogonal wavelets. Appl Comput Harmon Anal 1996;3(2):186
200.
[23] Jiang XQ, Blunt L, Stout KJ. Development of a lifting wavelet
representation for surface characterization. Proc R Soc Lond A
2000;456:2283313.

235

[24] Srinivasan V. Discrete morphological filters for metrology. In:

Proceedings of the 6th IMEKO ISMQC Symposium on Metrology
for Quality Control in Production. TU Wien, Austria, 1988.
[25] Scott PJ. Scale-space techniques. In: Proceedings of the Xth
International Colloquium on Surfaces. Chemintz University of
Technology, Chemnitz, 2000. p. 15361.