See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/44682747
Bilateral filter based orientation smoothing of
EBSD data
Article in Ultramicroscopy · June 2010
DOI: 10.1016/j.ultramic.2010.06.003 · Source: PubMed
CITATIONS
READS
3
39
2 authors:
Delphic Chen
Jui-Chao Kuo
20 PUBLICATIONS 119 CITATIONS
83 PUBLICATIONS 438 CITATIONS
National Cheng Kung University
SEE PROFILE
National Cheng Kung University
SEE PROFILE
All content following this page was uploaded by Jui-Chao Kuo on 24 June 2015.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Ultramicroscopy 110 (2010) 1297–1305
Contents lists available at ScienceDirect
Ultramicroscopy
journal homepage: www.elsevier.com/locate/ultramic
Bilateral filter based orientation smoothing of EBSD data
Delphic Chen, Jui-Chao Kuo n
Department of Materials Science and Engineering, National Cheng Kung University, 70101 Tainan, Taiwan
a r t i c l e in fo
abstract
Article history:
Received 31 October 2009
Received in revised form
24 May 2010
Accepted 1 June 2010
Bilateral filter based orientation smoothing was implemented in this study to improve the angular
precision of orientation maps for deposited and deformed structures of pure Cu obtained from electron
backscattered diffraction (EBSD) measurements. Applying the method to the deformed and deposited
structures, the accuracy of misorientation (or the limit of orientation noise) is enhanced from 0.71 to
0.251 and 0.071, respectively. Orientation smoothing has two features: preservation of boundary
structures or deformed substructures and significant reduction in orientation noise after only one pass.
& 2010 Elsevier B.V. All rights reserved.
Keywords:
EBSD
Orientation smoothing
Bilateral filter
Orientation noise
1. Introduction
Electron backscatter diffraction (EBSD) is widely used for
the quantitative characterization of crystallographic microstructures. Boundaries of low-angle misorientations that occur during
deformation are known as incidental dislocation boundaries [1].
Such misorientations in the subgrains may be in the range of
1–51, which is near the detectable angular resolution of 21 [2].
The detection of subgrain boundaries with misorientation
angles below 21 is then limited by the angular resolution.
According to Humphreys et al. [2], the angular resolution is
affected by a number of factors, such as the microscope operating
conditions, the pixels in the CCD camera, the resolution of the
digitized pattern, and the accuracy of the pattern solving
algorithms [3].
In subgrain boundaries the Kikuchi pattern quality becomes
worse with increase in strain because of both the overlapping
patterns in the vicinity of the subgrains and the increased
incoherent scattering associated with higher dislocation densities.
Rapid data collection using low-resolution electron backscatter
patterns (EBSPs) can also reduce the Kikuchi pattern quality. This
results in inaccurate orientations, which is considered the local
orientation deviation or orientation spread. As the EBSD data
collection rate increases, the local orientation spread, known as
the ‘‘orientation noise’’ [2,4], also increases and then the angular
resolution decreases.
Post-processing routines are important for reducing the
orientation noise and improving the angular accuracy. These
n
Corresponding author. Tel.: + 886 6 2757575x31130; fax: + 886 6 2754194.
E-mail address: jckuo@mail.ncku.edu.tw (J.-C. Kuo).
0304-3991/$ - see front matter & 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.ultramic.2010.06.003
routines are aimed at reducing the average orientation noise and
the number of artificial errors introduced by filtering. The
Kuwahara-based orientation filter was introduced by Humphreys
et al. [5] to reduce the average noise level. Godfrey [6] proposed a
modified Kuwahara filter to improve the edge preservation near
the triple junctions. Cho et al. [7] also introduced a new
orientation averaging method. The bilateral filter introduced by
Tomasi and Manduchi [8] can be used to smooth out image noise
and to retain the edge detail of an original image. This filter is now
applied in the field of image processing to recover details from
noisy signals. The bilateral filter, similar to a Gaussian filter,
smoothes the pixels in a digital image [9], but it does not smooth
the pixels if the color components (i.e. RGB value) are not similar.
An orientation smoothing method based on a bilateral filter is
used to reduce orientation noise and to preserve boundary
structures. Deposited and deformed (10% tensile strain) pure Cu
structures are applied to examine the performance of the
bilateral-based orientation smoothing method. The paper is
organized as follows: the concept of the orientation bilateral
filter is presented firstly in Section 2. Then the materials and the
experimental procedure are presented in Section 3. After that the
performance of the orientation bilateral filter is discussed in
terms of the reduction of the orientation noise and of the
preservation of the grain boundaries in Sections 4 and 5,
respectively. Finally, three filtering methods are compared in
Section 6.
2. Orientation bilateral filter
A simple Kuwahara filter, which uses the average orientation,
is used to remove the orientation noise in EBSD data. A conventional
1298
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
average orientation is obtained from the arithmetic mean value of
orientation data. Many possible methods have been proposed by
Lassen et al. [10], Humbert et al. [11], Humphreys et al. [5], and
Cho et al. [7] to determine the average orientation. Orientation
smoothing using Kuwahara filter involves replacing the central
point of interest in a square grid through the average orientation
with a minimum misorientation [5].
We propose a new orientation filter called ‘‘orientation
bilateral filter’’, which combines the average orientation [5] and
the bilateral filter [8] to preserve microstructural features and to
smooth orientation noise. The difference between the Kuwahara
filter and the orientation bilateral filter is that the latter uses the
weighted average of misorientations and the former uses the
arithmetic mean of the misorientations without considering
the influence of the distance from the point of interest. The
bilateral filter extends the concept of Gaussian smoothing by
weighing the filter coefficients (called a spatial weight function)
with their corresponding relative pixel intensities (called an
intensity weight function). The weighted average is computed
based on both the spatial and intensity weight functions for the
misorientation. The former and latter functions measure the
geometric distance and the misorientation between the center
point and the sampling point, respectively.
In addition to misorientations, indexing parameters, such
as the confidence index (CI), fit (or maximum angular deviation),
and image quality (IQ) are possible candidates for application
in the intensity weight function. The parameters of CI and fit
depend on the orientation of grains. Thus, it is difficult to directly
compare the value of these parameters for different orientations
as there exist no direct linking between the microstructure
and these parameters. The image quality maps display boundary
structures and features, which look like a dislocation network.
The average misorientation maps, in comparison with the image
quality, can also show boundary structures expressed in misorientation spreads between each orientation in a grain and the
average orientation. Therefore, the average misorientation was
chosen in this study to be the intensity weight function.
The smoothing process requires a quick computation of the
misorientation achieved using quaternion. The equation for
quaternion, which was first introduced by Hamilton [12,13], is
as follows:
Q ¼ cosðy=2Þ,
r1 sinðy=2Þ,
r2 sinðy=2Þ,
r3 sinðy=2Þ
Fig. 2. Effect of parameters sS and sR on the number of points with a
misorientation between 0.51 and 21 (orientation noise) for electrodeposited
copper.
ð1Þ
Fig. 1. Schematic illustration of the hexagonal sampling grids used for the
bilateral filtering process of a hexagonal image structure. The solid, dotted, and
dashed lines represent the hexagonal grids with 7, 19, and 37 sampling points,
respectively.
Fig. 3. Image quality map and (a) orientation maps for the cross-section of
electrodeposited copper using inverse pole figure color coding with the reference
direction in the normal direction (ND): (b) without a filtering process, (c) using a
hexagonal sampling grid of 7 sampling points, (d) using a hexagonal sampling grid
of 19 sampling points, (e) using a hexagonal sampling grid of 37 sampling points,
and (f) the inverse pole figure color coding. The black lines indicate grain
boundaries with misorientation angles between 0.51 and 62.81.
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
where r¼(r1,r2,r3) is the axis with a common numerical description for both the crystal and the reference axes, and y is the
rotation angle of the axis that completes the orientation
specification.
The average orientation for the center point, based on the
orientation bilateral filter, is computed by the weight function
n1
P
Q ðx0 ,y0 Þ ¼
Q ðxi ,yi Þ wðxi ,yi Þ
i¼0
n1
P
where Q(xi,yi) and Q(x0,y0) are the quaternions at the sampling
point (xi,yi) and the center point (x0,y0), respectively. The weight
function Q(xi,yi) is given by
wðxi ,yi Þ ¼ wS ðxi ,yi Þ wR ðxi ,yi Þ
2
wðxi ,yi Þ
i¼0
Fig. 4. Misorientation angle histogram of electrodeposited copper for raw data
and bilateral filtered data using 7, 19, and 37 sampling points. (The value of 62.8 is
referred to in [17].)
ð3Þ
The spatial weight function wS(xi,yi) and intensity weight
function wR(xi,yi) are obtained by
wS ðxi ,yi Þ ¼ e½ðxi x0 Þ
ð2Þ
1299
þ ðyi y0 Þ2 =2s2S
ð4Þ
and
2
wR ðxi ,yi Þ ¼ eðDgi0 Þ
=2s2R
ð5Þ
where sS is a parameter in the spatial domain, sR is a parameter in
the intensity domain, and Dgi0 is the misorientation between the
sampling point (xi,yi) and the center point (x0,y0).
The smoothing process is performed on a hexagonal grid of
measured orientations. A new orientation is replaced by the
averaged orientation, over a set of values at each hexagonal
grid point, according to Eq. (2). The size and shape of the entire
sampling region can have significant effects on the smoothing
process. The shape of the orientation bilateral filter is a hexagonal
grid that contains 37 points, including the central point.
These points correspond to the three nearest hexagonal neighborhoods (Fig. 1). Other grid sizes such as 7 (the nearest hexagon
neighbors) and 19 (two nearest hexagon neighbors) are also
discussed. The orientation bilateral filter varying the values of sS
and sR is applied on an EBSD data set to find the optimal values
for the spatial and intensity domain parameters sS and sR,.
The points with misorientations 0.5–21 are considered as the
orientation noise after the smoothing process. Fig. 2 shows
the effect of the parameters sS and sR on the number of the
orientation noise. The orientation noise for sS ¼ 2 and 4 is smaller
than that for sS ¼1 when the value of sR is in the range 0.5–5.
A minimum sR was observed at 2 when sS ¼2 and 4. Based on the
Fig. 5. Distribution of point-to-point misorientation angle for electrodeposited copper: (a) raw data; (b) with 7 sampling points; (c) with 19 sampling points; and (d) 37
sampling points. The Lorentz and Gauss fitting curves are shown as solid and dashed lines, respectively.
1300
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
observations, the parameters sS and sR are selected to be 2 and 2,
respectively.
Gaussian and Lorentz distribution functions are chosen to fit
the curve in order to investigate the distribution of misorientations 0–1.01. The Gaussian distribution is given by the formula
y ¼ y0 þ
2
A
2
pffiffiffiffiffiffiffiffi e2ðxxc Þ =w
w p=2
ð6Þ
where y0 is the offset, A the total area under the curve, xc the
mean, and w/2 the standard deviation, which is approximately
0.425 the full-width at half-maximum (FWHM).
The Lorentz distribution is given as
y ¼ y0 þ
2A
w
ð7Þ
p 4ðxxc Þ2 þ w2
where y0 is the offset, A the total area under the curve, xc the
mean, and w is FWHM.
Fig. 6. Maximum deviation after filter passes for electrodeposited copper.
3. Materials and experimental procedure
Cu film was synthesized through a pulsed electrodeposition
technique from an additive-free acidic Cu electrolyte 0.5 M CuSO4.
The pH value was adjusted to 1 by adding sulfuric acid.
Mechanical agitation was applied to ensure the refreshment of
the electrolytic solution and to prevent organic contamination. All
depositions were conducted at 20 1C under a current density of
0.5 A/cm2. The pulsed current chain was composed of time-on
0.2 s and time-off 2.0 s.
A pure Cu sample for the tensile test was prepared by an
electric discharge machine into the specimen dimensions. A
specimen was ground, polished with a 1 mm diamond paste, and
electropolished in a phosphoric-based electrolyte at 20 1C under
1.5 V for 60 s. This sample was deformed by 10% through microtensile testing (Kammrath and Weiss GmbH).
The deposited and deformed Cu samples were mechanically
polished through a standard metallographic procedure to a final
level of 0.03 mm. This was followed by electropolishing at a
voltage of 1.5 V for 60 s in a phosphoric acid electrolyte. EBSD
measurements were carried out using an EDAX/TSL Technology
EBSD system, including a Digiview IV Detector, mounted on a
field-emission scanning electron microscope (FESEM, JEOL JSM7001F). All EBSD data collection and analysis were performed
using TSLs OIM Data Collection and Analysis software 5.2. The
electron beam was operated at 30 kV with step sizes of 50 and
25 nm for the deposited and 10% strained Cu samples, respectively. To achieve a low orientation noise—that is, a high-quality
EBSD orientation map—the exposure time for the CCD camera
was set to 0.03 s for the deposited Cu and 0.05 s for the deformed
Cu. The binning size for the CCD camera was set to 4 4 and 2 2
pixels, for the deposited and deformed Cu, respectively.
Table 1
Fitting parameters of the average misorientation xc and the full-width at halfmaximum FWHM using Gauss and Lorentz functions for raw data and orientation
bilateral filtering with 7, 19 and 37 points in the case of deposit copper.
Sampling size
Raw data
7 points
19 points
37 points
Gauss function
Lorentz function
xc
FWHM
xc
FWHM
0.32,329
0.12,388
0.07,722
0.06,069
0.43,848
0.19,318
0.12,813
0.11,281
0.31,985
0.11,737
0.07,312
0.05,857
0.5005
0.18,023
0.11,464
0.09,729
Fig. 7. (a) Image quality map of pure copper after 10% tensile strain. Orientation
maps for the pure copper after 10% tensile strain using inverse pole figure color
coding with the reference direction in the normal direction (ND): (b) without the
filtering process, (c) using a hexagonal sampling grid of 7 sampling points, (d) using a
hexagonal sampling grid of 19 sampling points, (e) using a hexagonal sampling grid of
37 sampling points, and (f) the inverse pole figure color coding. The black lines
indicate grain boundaries with misorientation angles between 0.51 and 62.81.
1301
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
4. Orientation smoothing
4.1. Deposited structure
Fig. 3a shows the image quality map of a deposited structure of
Cu obtained using a scanning step size of 50 nm. The inverse pole
figure maps with highlighted boundaries obtained from the raw
ESBD data are shown in Fig. 3b. The black represents boundaries
with misorientations 0.5–62.81, which are comprised of the
orientation noise of 0.5–51, low-angle grain boundaries (LAGBs) of
misorientations 5–151, and high-angle grain boundaries (HAGBs) of
misorientations 15–62.81 [17]. The ratio of 62,926 points, with
misorientations 0.5–51, to the total measured number of 92,220
points was 68.3% (Fig. 4).
Three sampling hexagonal grids containing 7, 19, and 37 points
were chosen to filter the raw EBSD data during orientation
smoothing (Fig. 1). The number of the orientation noise was
significantly reduced from 62,926 to 10,835, 4806, and 3,660
(corresponding reductions from 68.3% to 11.7%, 5.2%, and 3.9%) for
the points 7, 19, and 37, respectively (Figs. 3c and 4). As the
number of sampling points increases, the number of the
orientation noise decreases. The number of LAGBs and HAGBs
remains unchanged after filtering (Fig. 4).
The angular resolution is considered as a lower limit of
misorientation between points recognized above the noise
threshold. According to Humphreys et al. [2], the EBSD angular
resolution, which depends on the EBSD detector, cell size, and
beam conditions, approaches 11. The misorientation distribution
of the raw data is displayed from 01 to 1.01 in Fig. 5a. Lorentz
function gives a better fit for the range of 0.6–1.01 compared with
Table 2
Fitting parameters of the average misorientation xc and the full-width at halfmaximum FWHM using Gauss and Lorentz functions for raw data and orientation
bilateral filtering with 7, 19, and 37 points in the case of deformed copper.
Sampling points
Fig. 8. Misorientation angle histogram of pure copper after 10% tensile strain for
raw data and after bilateral filtering using 7, 19, and 37 sampling points.
(The value of 62.8 is referred to in [17].)
Raw data
7
19
37
Gauss function
Lorentz function
xc
FWHM
xc
FWHM
0.32,046
0.15,468
0.11,792
0.11,172
0.37,553
0.25,067
0.21,449
0.20,623
0.31,572
0.14,701
0.11,166
0.10,596
0.40,830
0.24,110
0.20,010
0.19,115
Fig. 9. Distribution of point-to-point misorientation angle for pure copper after 10% tensile strain: (a) raw data, (b) with 7 sampling points, (c) with 19 sampling points, and
(d) with 37 sampling points. The Lorentz and Gauss fitting curves are shown as solid and dashed lines, respectively.
1302
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
Fig. 10. Maximum deviation after filter passes for deformed copper.
the Gauss function (Fig. 5a). However, both functions fail to fit the
raw data in the range of 0–0.251. The computed Gaussian function
fitting indicates an average misorientation of 0.3231 with an
FWHM of 0.4381, while a mean misorientation of 0.3201 with an
FWHM of 0.5011 is obtained using the Lorentz function (in
Table 1). The value of the angular resolution here is derived by
adding the mean value and two standard deviations (2s), which is
approximated by 0.849 FWHM. Therefore, an angular resolution
of 0.71 is achieved in the EBSD raw data without filtering.
The effects of the sampling size (the number of the sampling
points) and the filter pass number on the reduction of the
orientation noise using the orientation bilateral filter are
investigated. The limits of the orientation noise are 0.291, 0.191,
and 0.161 (reducing them by factors of 2.4, 3.7, and 4.4) after one
pass using 7, 19, and 37 points, respectively. The reduction factors
were 4.8, 8.2, and 9.8, corresponding to the orientation noise
values of 0.15, 0.08, and 0.07 after five passes using 7, 19, and 37
points, respectively (Fig. 6). The decrease in the orientation noise
is not significant when the number of the sampling points is
greater than 19. If the pass number is 45, the angular resolution
Fig. 11. ODF sections with the constant j2 using the Bunge angular notation for electrodeposited copper: (a) without the filtering process; (b) using a hexagonal sampling
grid of 7 sampling points; (c) using a hexagonal sampling grid of 19 sampling points; and (d) using a hexagonal sampling grid of 37 sampling points. The intensity scale is
given in (f).
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
approaches to the limit value of 0.071, which is recognized as the
angular resolution with filtering. The angular resolution
after filtering is clearly improved from an initial value of 0.71
to 0.071.
4.2. Deformed substructure
The image quality map taken a step size of 25 nm and having
65,008 total measured points shows the 10% tensile strained
structure of pure Cu (Fig. 7a). The subgrain structure is clearly
seen in the image quality map. The corresponding map using
color coding in terms of an inverse pole figure is overlaid with
reconstructed boundaries (Fig. 7b). The black represents
boundaries with misorientations 0.5–62.81, which include the
orientation noise (0.5–51), LAGBs (5–151), and HAGBs (15–62.81).
Fig. 7c–e illustrates that applying this technique to the ESBD
orientation data significantly reduces the amount of the orientation noise. The fraction of the orientation noise is significantly
decreased to 28.7% of the total measured points after one pass
using 37 points (Fig. 8).
1303
An average misorientation of 0.3201 with an FWHM of 0.3761
is determined using the Gaussian function for the raw data, while
a mean misorientation of 0.3161 with an FWHM of 0.4081 is
obtained using the Lorentz function (Fig. 9 and Table 2).
An angular resolution of 0.71 is determined for the deformed
Cu structure without filtering, based on the definition in
Section 4.1. The angular resolution without filtering is 0.71 for
the cases of deposited and deformed structures. A limit value
of 0.71 is then considered as the angular resolution for FESEM in
this study.
The orientation noise limit is decreased to 0.281, 0.291, and
0.281 after orientation smoothing using 7, 19, and 37 points with
one pass, respectively (Fig. 10). With five passes, the values of the
orientation noise using 7, 19, and 37 points are 0.291, 0.261, and
0.251, respectively. The decrease in the orientation noise is not
significant when the number of the points is greater than 19.
If the pass number is 42, the angular resolution approaches to a
limit value of 0.251 after filtering. The noise limit of 0.251,
determined for the deformed structure in Fig. 10, is smaller than
that of 0.071 for the deposited structure in Fig. 6. This observation
suggests that increase in the orientation noise from 0.071 to 0.251
Fig. 12. ODF sections with a constant j2 using the Bunge angular notation for pure copper after 10% tensile strain: (a) without the filtering process; (b) using a hexagonal
sampling grid of 7 sampling points; (c) using a hexagonal sampling grid of 19 sampling points; and (d) using a hexagonal sampling grid of 37 sampling points.
1304
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
is due to the formation of subgrain structures with LAGBs during
deformation.
5. Preservation of boundary structure
5.1. Deposited structure
The preservation of boundary structures is now presented here
after discussing the decrease in the orientation noise. The
orientation distribution function (ODF) sections demonstrate the
preservation of boundary structures, including LAGBs and HAGBs.
These ODF sections reveal a similar orientation distribution as
that obtained from the raw EBSD data after smoothing with 7, 19,
and 37 points (Fig. 11). However, the maximum intensity in the
ODF sections increases as the number of the points increases from
7 to 37. This observation of increase in orientation intensity
results from the reduction in the orientation noise and
preservation of the initial orientation (Fig. 4).
The frequency of LAGBs and HAGBs after orientation smoothing is compared with the raw data in Fig. 4. The number of LAGBs
and HAGBs remains unchanged after smoothing, as in the case of
the raw data after smoothing. Therefore, orientation smoothing
leads to a drastic reduction in the orientation noise.
5.2. Deformed substructure
The 10% tensile strained structure after orientation smoothing
is illustrated in Fig. 7. These ODF sections, using 7, 19, and 37
points, reveal a similar orientation distribution as the initial ODF
sections in the EBSD raw data, as shown in Fig. 12. However, the
maximum intensity in the ODF section increases as the number of
sampling points increases from 7 to 37, showing the same trend
as that of the deposited structure.
The deformed structure consists of LAGBs, HAGBs, and the
orientation noise (Fig. 8). The frequencies of LAGBs and HAGBs
remain unchanged, similar to the raw data after smoothing.
Deformed structures typically contain cells or subgrains with
misorientations 0.5–31 [14–16]. The boundaries with misorientations 0.5–5.01 made up a large fraction (28%) of the deformed
structure after orientation smoothing. This is larger than 4%
observed for the deposited structure. Subgrain boundaries with
misorientations between 0.51 and 5.01 are formed during
deformation.
6. Comparison of Kuwahara filtering, orientation bilateral
filtering, and cleanup processing
The performances of the orientation bilateral and the Kuwahara filters are compared in terms of their orientation noise
reduction and the boundary preservation of deposited and
deformed structures. A symmetrical square grid of 5 5 pixels
was adopted for the Kuwahara filtering and the minimum
misorientation was selected as 21. A hexagonal grid containing
19 sampling points was chosen for the orientation bilateral filter,
and other parameters were maintained similar to those seen in
Section 2. As well, the built-in cleanup function of the TSL OIM
Analysiss 5.1 for a single orientation per grain was also used as a
reference. This process is called the cleanup process. The
minimum misorientation angle for the cleanup process was also
set to 21. The TSL OIM analysis software was used to analyze all
EBSD data after Kuwahara and orientation bilateral filtering.
The boundaries are classified into four groups, based on the
following misorientations: 0.5–2.01 (purple), 2.0–5.01 (red),
5.0–15.01 (black), and 15.0–62.81 (green). The image quality maps
Fig. 13. Image quality maps for the deposited copper: (a) without the filtering
process; (b) using ‘‘single orientation per grain’’ cleanup method; (c) using
Kuwahara filter; (d) using orientation bilateral filter of 19 sampling points. The
purple, red, black, and green lines indicate grain boundaries with misorientation
angles between 0.51 and 21, 21 and 51, 51 and 151, and 151 and 62.81, respectively.
(For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)
Table 3
Number of boundary misorientations of 0.5–21, 2–51, 5–151, and 15–62.81 for raw
data, cleanup processing, Kuwahara filtering, and orientation bilateral filtering,
respectively, in the case of deposited copper.
Filtering method
0.5–2o
2–5o
5–15o
15–62.8o
Raw
Cleanup
Kuwahara
Bilateral
65,529
14,438
33,105
8717
633
7765
534
436
371
372
370
372
25,902
25,903
25,902
25,904
Table 4
Number of boundary misorientations of 0.5–21, 2–51, 5–151, and 15–62.81 for raw
data, cleanup processing, Kuwahara filtering, and orientation bilateral filtering,
respectively, in the case of deformed copper.
Filtering method
0.5–21
2–51
5–151
15–62.81
Raw
Cleanup
Kuwahara
Bilateral
50,053
12,903
36,255
16,878
4307
8580
3953
4095
2291
7721
2450
2467
1942
1936
614
602
with boundaries are shown in Fig. 13 for the deposited Cu
structure. Boundaries with misorientations 0.5–21 contain 65,529
points corresponding to 68.8% of the total measured points in the
raw data (Table 3). The number of the points with misorientations
0.5–2.01 is reduced to 33,105 (35.9%), 8717 (9.4%), and 14,438
(15.7%), respectively, after one pass using Kuwahara filtering,
orientation bilateral filtering, and cleanup processing (as shown in
Table 3). The number of boundaries with misorientations 2.0–5.01
increases after cleanup processing (Table 4). The number of
D. Chen, J.-C. Kuo / Ultramicroscopy 110 (2010) 1297–1305
1305
deposited and deformed structures. This orientation smoothing
provides a simple and rapid alternative to reducing the orientation noise and preserving boundary structures after a single pass.
The angular resolution of the measured EBSD data without
filtering is 0.71, which is determined from the limit of the
orientation noise for deposited and deformed Cu structures. The
angular resolution is improved from the initial value of 0.71 to
0.071 and 0.251 after the application of the filter to deposited and
deformed structures of pure Cu, respectively.
Acknowledgements
The authors would like to thank the National Science Council
of the Republic of China under Contract no. NSC 98-2221-E-006 081 -MY2, and the Center for Micro/Nano Science and Technology
for providing facilities.
References
Fig. 14. Image quality maps for the pure copper after 10% tensile strain: (a)
without the filtering process; (b) using ‘‘single orientation per grain’’ cleanup
method; (c) using Kuwahara filter; (d) using orientation bilateral filter of 19
sampling points. The purple, red, black, and green lines indicate grain boundaries
with misorientation angles between 0.51 and 21, 21 and 51, 51 and 151, and 151 and
62.81, respectively. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
boundaries at 2.0–5.01, 5.0–15.01, and 15.0–62.81 remains almost
the same for both Kuwahara and orientation bilateral filtering.
Fig. 14 shows image quality maps with boundaries for the 10%
strained pure Cu. The fraction of points with misorientations
0.5–2.01 is 77.0% (50,053 points) in the case of the raw data. This
is larger than 68.8% of the deposited Cu. This could be due to the
formation of subgrain boundaries during straining. The number of
the points with misorientations 0.5–2.01 is reduced to 36,255
(55.7%), 16,878 (25.9%), and 12,903 (20%), respectively, after
applying Kuwahara filtering, orientation bilateral filtering, and
cleanup processing. The number of boundaries with
misorientations of 2.0–5.01 and 5.0–15.01 is increased using
cleanup processing (Table 4), while the number of boundaries at
2.0–5.01, 5.0–15.01, and 15.0–62.81 remains unchanged after
application of the Kuwahara and the orientation bilateral filtering.
Therefore, compared with the Kuwahara filter, the orientation
bilateral filter provides an alternative and more rapid method of
correcting the orientation noise and preserving the boundaries
after one pass filtering.
7. Conclusions
A bilateral filter based orientation smoothing was successfully
applied to increase the angular resolution of the EBSD data for
[1] D.A. Hughes, Q. Liu, D.C. Chrzan, N. Hansen, Scaling of microstructural
parameters: misorientations of deformation induced boundaries, Acta Mater.
45 (1997) 105–112.
[2] F.J. Humphreys, Y. Huang, I. Brough, C. Harris, Electron backscatter diffraction
of grain and subgrain structures: resolution considerations, J. Microsc. 195
(1999) 212–216.
[3] D.J. Dingley, V. Randle, Microtexture determination by electron back-scatter
diffraction, J. Mater. Sci. 27 (1992) 4545–4566.
[4] F.J. Humphreys, Y. Huang, I. Brough, C. Harris, High resolution electron
backscatter diffraction with a field emission gun scanning eectron
microscope, J. Microsc. 195 (1999) 4–10.
[5] F.J. Humphreys, P.S. Bate, P.J. Hurley, Orientation average of electron
backscattered diffraction data, J. Microsc. 201 (2001) 50–58.
[6] A. Godfrey, Edge preservation near triple junctions during orientation
averaging of EBSP data, Scr. Mater. 50 (2004) 1097–1101.
[7] J.H. Cho, A.D. Rollet, K.H. Oh, Determination of a mean orientation in electron
backscatter diffraction measurements, Metall. Mater. Trans. A 36A (2005)
3427–3438.
[8] C. Tomasi, R. Manduchi, Bilateral filtering for gray and color images, in:
Proceedings of the Sixth International Conference on Computer Vision
(ICCV’98), IEEE, USA, pp. 839–846.
[9] L.G. Shapiro, G.C. Stockman, Computer Vision, Prentence Hall, NJ, USA, 2001.
[10] N.C.K. Lassen, D.J. Jensen, K. Conradsen, On the statistical analysis of
orientation data, Acta Crystallogr. A 50 (1994) 741–748.
[11] M. Humbert, N. Gey, J. Muller, C. Esling, Determination of a mean
orientations: applications to electron back-scattering pattern measurements,
J. Appl. Crystallogr. 29 (1996) 662–666.
[12] P.V. Arunachalam, W. R. Hamilton and his quaternions, Math. Educ. 6 (1990)
261–266.
[13] S.L. Altmann, Rotations, Quaternions and Double Groups, Clarendon Press,
Oxford, UK, 1986.
[14] F.J. Humphreys, M. Hatherly, Recrystallisation and Related Annealing
Phenomena, Elsevier, Oxford, UK, 1995.
[15] B. Bay, N. Hansen, D.A. Hughes, D. Kuhlman-Wilsdorf, Evolution of FCC
deformation structures in polyslip, Acta Metall. Mater. 40 (1992) 205–219.
[16] N. Hansen, D.J. Jensen, Development of microstructure in fcc metals during
cold work, Philos. Trans. R. Soc. London Ser. A Math. Phys. Eng. Sci. 357 (1999)
1447–1469.
[17] J.K. MacKezie, Second paper on statistics associated with the random
disorientation of cubes, Biometrica 45 (1958) 229–240.