DigitalMicrograph

Multivariate Histogram
Analysis User’s Guide

Gatan, Inc.
5933 Coronado Lane
Pleasanton, CA 94588
Tel (925) 463-0200
FAX (925) 463-0204

May 2001

Revision 1
 Preface

About this Guide

This Multivariate Histogram Analysis User’s Guide is written to provide
procedure for the calculation and analysis of multivariate histograms within
DigitalMicrograph. This Guide assumes the user is familiar with image
manipulation within DigitalMicrograph and only addresses those features
specific to the Multivariate Histogram Analysis package.

Preview of this Guide

The Multivariate Histogram Analysis User’s Guide includes the following
chapters:
Chapter 1, “Introduction” summarizes the purpose and applicability of the
Multivariate Histogram Analysis software and notes the software and hardware
requirements.
Chapter 2, “Performing Multivariate Histogram Analysis” provides an
overview to the multivariate histogram calculation and analysis process using
DigitalMicrograph.

Multivariate Histogram Analysis User’s Guide Rev 1 i

 Disclaimer
Gatan, Inc., makes no express or implied representations or warranties with
respect to the contents or use of this manual, and specifically disclaims any
implied warranties of merchantability or fitness for a particular purpose. Gatan,
Inc., further reserves the right to revise this manual and to make changes to its
contents at any time, without obligation to notify any person or entity of such
revisions or changes.

Copyright and Trademarks

© 2001. All rights reserved.
DigitalMicrograph ® is a registered trademark of Gatan. Inc., registered in the
United States.

ii Multivariate Histogram Analysis User’s Guide Rev 1

 Support

Contacting Gatan Technical Support

Gatan, Inc., provides free technical support via voice, Fax, and electronic mail.
To reach Gatan technical support, call or Fax the facility nearest you or contact
by electronic mail:
• Gatan, USA (West Coast)
Tel: (925) 463 0200
Fax: (925) 463 0204
• Gatan, USA (East Coast)
Tel (724) 776 5260
Fax: (724) 776 3360
• Gatan, Germany
Tel: 089 352 374
Fax: 089 359 1642
• Gatan, UK
Tel: 01865 253630
Fax: 01865 253639
• Gatan, Japan
Tel: 0424 38 7230
Fax: 0424 38 7228
• Gatan, France
Tel: 33 (0) 1 30 59 59 29
Fax: 33 (0) 1 30 59 59 39
• Gatan, Singapore
Tel: 65 235 0995
Fax: 65 235 8869
• Gatan Online
http://www.gatan.com/
[email protected]
[email protected]

Multivariate Histogram Analysis User’s Guide Rev 1 iii

iv Multivariate Histogram Analysis User’s Guide Rev 1
 Table of Contents

Preface i
Support iii
Table of Contents v
List of Figures vii
1 Introduction 1-1
1.1 Multivariate Histogram Analysis Summary 1-1
1.2 Requirements 1-2
1.2.1 Microscope Requirements 1-2
1.2.2 Computer Requirements 1-2
1.3 Software Installation 1-2
2 Performing Multivariate Histogram Analysis 2-1
2.1 Setting up the histogram parameters 2-1
2.2 Creating a bivariate histogram plot 2-3
2.3 Creating a trivariate histogram plot 2-5
2.4 Extracting a single phase image from a histogram plot 2-8
2.5 Extracting a composite phase image 2-10
2.6 Extracting a single phase image from a composite map 2-13
2.7 Performing phase-specific spectroscopy on a spectrum-image 2-14
2.8 Advanced notes 2-15

Index I-1

Multivariate Histogram Analysis User’s Guide Rev 1 v

vi Multivariate Histogram Analysis User’s Guide Rev 1
 List of Figures

Figure 2-1 The



 
  sub-menu. 2-1
Figure 2-2 The Histogram Setup dialog. 2-2
Figure 2-3 Source images for the bivariate histogram in Figure 2-4. 2-4
Figure 2-4 A bivariate histogram calculated from the two source images shown in
Figure 2-3. 2-5
Figure 2-5 Source images for the trivariate histogram in Figure 2-6. 2-6
Figure 2-6 Trivariate histogram (with annotations) of the source images shown in
Figure 2-5. 2-7
Figure 2-7 Extracting a phase-specific map (right) from a bivariate histogram
(left). 2-8
Figure 2-8 Extracting a phase-specific map (right) from a trivariate histogram. 2-9
Figure 2-9 Generating a composite phase image (bottom) from a bivariate histogram
(left). The color-encoded ROI map is shown right with ICP palette
overlaid. 2-10
Figure 2-10 The composite phase image (right), with corresponding color encoded
ROI map (left). 2-11
Figure 2-11 Composite phase image (right) with corresponding trivariate ROI map
(left). 2-12
Figure 2-12 Single phase map extraction from a composite phase image (left) using
the EXTRACT SINGLE PHASE MAP FROM COMPOSITE option. 2-13
Figure 2-13 Initiating phase-specific spectroscopy on an EELS spectrum-image (left)
using a phase-image (right). 2-14
Figure 2-14 Phase specific spectrum generated as shown in the previous figure. 2-15
Figure 2-15 The Multivariate Histogram Analysis preference tags. 2-16

Multivariate Histogram Analysis User’s Guide Rev 1 vii

viii Multivariate Histogram Analysis User’s Guide Rev 1
 1 Introduction

Multivariate histogram analysis, also known as scatter plot analysis, provides

an effective mechanism for exploring the spatial correlations of the intensity
distributions of multiple spatially aligned images. When the source images are
chemical distribution images, for example EELS jump-ratio or background
extrapolated maps, the intensity correlations within the resultant multivariate
histogram reveals information regarding the spatial correlation of the
represented elemental species. Discrete, or semi-discrete, ‘clusters’ of
intensity within the histogram plots indicate the presence of phases with distinct
chemical composition or stoichiometry. Isolating these clusters and retrieving
their corresponding spatial representations, a process termed ‘intercorrelative
partitioning’, yields chemical-phase specific maps. Such maps can reveal
unique information regarding the spatial distribution of the chemical phases
present, can provide a convenient means for data reduction, and can also be
used as spatially selective masks allowing phase-specific spectroscopy to be
performed on spectrum-image datasets.

1.1 Multivariate Histogram Analysis Summary

A number of tools for the calculation and analysis of multivariate histograms

are contained in the "!$#&%('*)+,-'*+%(./0'*12%(345,-+687:9$+#&;(12'*1
hierarchical sub-menu,
located within the 7:9$+#&;(12'*1
menu. These tools enable both bivariate and
trivariate histograms to be generated, using two or three spatially aligned source
images, respectively. Utilities for partitioning histogram regions, and trace-back
algorithms for the retrieval of the corresponding chemical phase maps, are also
included, as well as commands for performing phase-specific spectroscopy on
spectrum-image datasets. This tutorial will describe the use of each sub-menu
item in the context of both bi- and trivariate histogram analysis, describing in
detail the routine procedures where appropriate. For an introduction on the
principles and practical application of histogram analysis of chemical specific
maps, please refer to the following reference:
Grogger W., Hofer F. and Kothleitner G., “Quantitative chemical phase analysis of EFTEM elemental
maps using scatter diagrams”, Micron 29 (1998) 43 -51

Multivariate Histogram Analysis User’s Guide Rev 1 1-1

 1.2 Requirements

1.2.1 Microscope Requirements

No microscope is required for performing multivariate histogram analyses.
However, a microscope equipped with appropriate hardware is required for the
acquisition of the datasets to be analyzed. Please refer to the appropriate
hardware documentation for details.

1.2.2 Computer Requirements

Multivariate Histogram Analysis runs as part of the Gatan Microscopy Suite
(GMS), and hence the minimum and recommended computer requirements are
as specified for running the GMS. Please refer to the GMS Installation Guide
for specific details.

1.3 Software Installation

The Multivariate Histogram Analysis software plug-in is installed as part of the

Gatan Microscopy Suite (GMS). Providing your Gatan Software License Disk
contains an appropriate license configuration, the Multivariate Histogram
Analysis software plug-in will be configured automatically on installation of
the GMS. Please refer to the GMS Installation Guide for installation details.

1-2 Multivariate Histogram Analysis User’s Guide Rev 1

 2 Performing Multivariate
Histogram Analysis

This section gives a step-by-step guide to generating and using multivariate

histogram plots within the context of analyzing multiple EELS or energy-
filtered TEM chemical maps. The principles described, however, can equally
be applied to other types of imaging, for example, EDX mapping, or indeed to
<"=$>&?(@*ABC-@*B?(DE0@*F2?(GH5C-BI8J:K$B>&L(F2@*F
any images with complementary intensity distributions. The menu items in the
sub-menu, shown in Figure 2-1, are
ordered sequentially in the logical order one would normally follow when
performing such analyses. Accordingly, the following sub-sections are ordered
sequentially following the menu structure.

Figure 2-1 The MONQP&R&S*TVUVWS*UVR&XZYS*[2R&\Q]-WUV^_Z`5UVP&ab[2S*[ sub-menu.

2.1 Setting up the histogram parameters

The c0d*e2f(gh5i-jkmlbnVf(o$pq

item initiates the Histogram Setup dialog, as shown in
Figure 2-2. This dialog allows a number of the basic properties of the
generated two or three-dimensional histograms to be defined, namely histogram
resolution and display properties. The dialog contains the following parameter
fields;

Multivariate Histogram Analysis User’s Guide Rev 1 2-1

 Setting up the histogram parameters

Histogram resolution (pixels) - This value determines the resolution (in pixels)
of the generated histogram. In the case of a bivariate histogram, the generated
plot will be square with side length equal to this value. In the trivariate case,
the resultant histogram will be an equilateral triangle of base length equal to
this value. The optimal histogram resolution will depend on both the intensity
distribution and the quantity of pixels within the source images. A lower
resolution is favored for source images containing a low number of pixels, and
vice-versa. Lowering the histogram resolution will produce a coarser looking
histogram. Raising this value will increase the sampling density, but at the
expense of mean counts per pixel within the histogram.

Figure 2-2 The Histogram Setup dialog.

Source survey technique – This pop up menu contains the available intensity
survey techniques for determining the minimum and maximum intensities of
the source images that will be considered in the histogram generation process.
Hence, the technique selected in this field determines the intensity range of
each histogram axis. The first four survey methods, cross-wire, whole image,
sparse and reduction, are the default contrast survey methods provided and
used by Digital Micrograph; please refer to your Digital Micrograph manual for
a full account of the surveying criteria these techniques apply. Selecting one of
these survey methods will result in an automatic determination of the histogram
intensity range, and experimentation is often required to establish the most
suitable for a particular dataset. The fifth survey technique, source display,
applies the intensity range of the source images as displayed in their respective
image-windows. Hence, by adjusting the source image-display intensity ranges
using, for example, the r0s*t2u(vw5x-yz
floating window, the minimum and
maximum values can be specified giving the user complete control over the
histogram range.
Histogram color table – This option determines the color table used for the
multivariate histogram display. The available options are taken from the Digital
Micrograph default color tables; the choices are grayscale and temperature.
Composite phase-map color table – The choice in this pop-up menu determines
the color table applied to the generated composite phase-maps, described later
in this section. The color tables are based on the identically titled Digital
Micrograph defaults, with the choice of grayscale, temperature and rainbow.

2-2 Multivariate Histogram Analysis User’s Guide Rev 1

 Creating a bivariate histogram plot

The default values will give satisfactory results in the majority of cases, though
it is recommended you become familiar with the effects of these parameters,
particularly the histogram resolution and survey technique settings, in order to
optimize your histogram plots. Further histogram display attributes may also be
adjusted; please refer to the advanced notes in Section 2.8for further details.

2.2 Creating a bivariate histogram plot

Selecting the {Q|-}V~(}€5*‚~|-*~(}ƒ0*„2(…†5|-~‡

sub-menu item initiates the
bivariate histogram calculation. The routine proceeds by requesting you to
specify two source images, which should be spatially aligned and also have
dimensions of equal sizes. If the two source images are spatially misaligned,
the accuracy of the generated histogram will be compromised. Alternatively, if
the images have incompatible dimensionalities then the procedure will be not
proceed and you will be alerted with a suitable error message. In either case,
ensure that the above criteria are met by cropping the source images if
necessary.
Once the source images have been specified, the procedure determines for each
the intensity range to be considered in the histogram calculation, applying the
specified survey method as described in the previous section. Once the
minimum and maximum intensities are established, the histogram is generated
using a sampling width w determined by the histogram resolution r (as
specified in the histogram setup dialog) following, for an image A,

max survey ( A) − min survey ( A)

wA =
r
Please note that the sampling density is calculated for each axis of the
histogram, resulting in an equal number of samples for each source image and,
hence, a square histogram. Additionally, note that pixels with intensity values
outside the survey range are not excluded from the histogram analysis; these
pixels are clipped to have values at the corresponding intensity limit, and will
therefore be included in the histogram at the extremes of the range covered.
The histogram is generated by comparison of the intensities contained within
the two source images subject to the minimum and maximum values (as
determined by the survey technique specified) and the corresponding sampling
width (governed by the histogram resolution). Each pixel in source image A
will have a corresponding pixel at the same spatial co-ordinate (x,y) in source
image B. Representing the associated pixel intensities as a and b for source
images A and B respectively, each corresponding pixel pair will have a
contribution of unity to the histogram at co-ordinate (a, b). The resultant value
of the histogram at co-ordinate (a, b) therefore represents the total number of
pixel pairs in A and B with intensities a and b respectively, within the sampling
widths wa and wb defined as above.

Multivariate Histogram Analysis User’s Guide Rev 1 2-3

 Creating a bivariate histogram plot

Figure 2-3 Source images for the bivariate histogram in Figure 2-4.

An example of creating a bivariate histogram is described as follows. The two

source images, shown in Figure 2-3, are chemical distribution maps taken from
a sample consisting of gold particles in on an amorphous carbon support film.
The carbon and gold maps shown were calculated using the jump-ratio method,
from energy-filtered images taken across the C K and Au N edges respectively.
The procedure for generating a bivariate histogram is as follows:

To create a bivariate histogram

1. Ensure the two source images are of similar size and are spatially
aligned. If the images are incompatible sizes, they will have to be
cropped to a common size. To check if the images are spatially aligned,
perform a cross-correlation (e.g. using the ˆ‰QŠ$‹ŒŽ‰Z‰Q‘Š$’“•”$‰–Ž“•—’
option in the ˜$“•‘’‰QŽ
menu). If the images are not spatially aligned,
then they should be cropped accordingly. (Hint: to crop an image,
place a rectangular ROI on the region of interest, copy the contents
using Ctrl +C, then paste the region into a new image window using
Ctrl+Alt+V.)
2. Enter the desired options in the histogram setup dialog, initiated by
selecting the ™š“•‹’›œŽŠ$ž Ÿ5‰Q’Œ¡
sub-menu item. For the example, a
histogram resolution of 256 pixels was specified (a resolution
comparable to the source image side length), the Source Display survey
technique was selected, and Temperature histogram color table was
used.
3. If the Source Display survey technique is selected, specify the
maximum and minimum intensity levels by either highlighting the
desired range using the Histogram floating window, or by entering the
values directly in the appropriate fields via dialog the ¢£ž¤Š$œ‰
–“•‹¡-‘Š$¥¦¨§
dialog option found under . ©0ª$«&‰Q¬Q’

2-4 Multivariate Histogram Analysis User’s Guide Rev 1

 Creating a trivariate histogram plot

4. Select ­ ®¯Q°$±¯²³•´$°$®³•°$±¯µš³•¶±·¸®°$¹ from the º»¼±³•´$°$®³•°$±¯

µš³•¶±·¸®°$¹m½¿¾°$¼À5¶³•¶ sub-menu, and specify the two source images in
the pop-up menu dialog.
5. The bivariate histogram will be calculated and displayed in a new
image window. If the displayed intensity range is not to your liking,
repeat steps 2-4 using a different survey technique or alternatively, if
the Source Display survey technique is specified, select a more
appropriate intensity range.

Figure 2-4 A bivariate histogram calculated from the two source images shown in Figure 2-3.

The calculated histogram is shown in Figure 2-4. It is sometimes possible for a

few intense regions to dominate the histogram image display intensity range,
causing regions of low value to appear black. If this happens, adjust the image-
display intensity range as appropriate. For a bivariate histogram calculated in
this way, the axes are linear and represent the sampled intensity range of the
source images in their calibrated intensity units. The value of the pixel at the
intersection of the two red lines gives the number of pixels in the Au and C
maps with jump-ratio values of 0.59 and 1.7 respectively. It should be noted
that the histogram shown consists of two well-defined clusters, with a relatively
high carbon jump-ratio and the other with a higher gold jump-ratio, joined by
an intermediate branch of intensity. As will be demonstrated in Section 4.5,
this is indicative of two well-defined chemical phases / species, with a small
degree of spatial overlap.

2.3 Creating a trivariate histogram plot

A trivariate histogram may be generated from three spatially aligned images

following a similar procedure as described in the previous section for a
bivariate histogram. In the same way that a one-dimensional histogram is

Multivariate Histogram Analysis User’s Guide Rev 1 2-5

 Creating a trivariate histogram plot

represented by a line-plot and a bivariate histogram may be represented as a

rectangular plot, a trivariate histogram can be represented as a cuboid.
However, manipulating a three-dimensional form on a two-dimensional
computer screen poses many problems in terms of visualization and
manipulation, especially when performing intercorrelative partitioning of
phases as described later in Section 2.4. A more practical and intuitive way of
performing trivariate histogram analysis is by means of a triangular histogram
in a similar style to a tertiary phase diagram. In the case where an equal
number of samples are taken of the source images the resultant trivariate
histogram plotted in this way is an equilateral triangle.

Figure 2-5 Source images for the trivariate histogram in Figure 2-6.

A tertiary phase diagram can be displayed as a two-dimensional plot since the

sum of the (three) constituents always adds up to 100%. However, this is not
necessarily the case for a trivariate histogram plot from three chemical
distribution maps; there could well be other constituent elements present, or
alternatively the source images may be generated using a qualitative, but not
quantitative, mapping technique such as jump-ratio imaging. To accommodate
this, the intensities are plotted as a fraction F of the total sum of intensities for
each pixel from the three source images A, B and C following

a
Fa =
a+b+c
where a is the pixel intensity at A(x, y) and so on. Note that this normalization
is performed after applying the intensity range limits to the source images
following the specified survey technique (as described for bivariate analysis),
and after rescaling with the minimum intensity set at zero and the maximum at
the specified histogram resolution – 1. This rescaling allows the contribution of
each source image to be independent of its relative intensity, avoiding maps
with high counts dominating the histogram. Hence the computed trivariate
histogram does not have linear axes and the plot cannot be regarded to be
strictly quantitative as such (to achieve this, a cuboid histogram would have to
be used). As a consequence, the axes are not marked with a scale. Therefore
the distribution of intensities in the trivariate histogram should be regarded as a
qualitative representation of the quantitative distribution, i.e. indicative of the

2-6 Multivariate Histogram Analysis User’s Guide Rev 1

 Creating a trivariate histogram plot

spatial correlation between the constituent images, where clusters of intensity

indicate a phase may be present.
An example of a trivariate histogram calculation is described as follows. The
three source images are shown in Figure 2-5. These maps were taken from an
EELS spectrum-image acquired from a semi-conductor device sample, and
calculated using the background extrapolation method for mapping. The
procedure for calculating a trivariate histogram is almost identical to the
previous procedure for calculating a bivariate histogram;

To create a trivariate histogram

1. Ensure the three source images are spatially aligned and are of equal
size. If not, crop accordingly as described in Step 1 above in the
bivariate histogram example.
2. Set the desired histogram setup options and intensity ranges (if
necessary) following steps 2 and 3 as described above for bivariate
histogram calculation.

È Ã•ÉÂʝËƏÅ$ÌÎÍ¿ÏÅ$ÁÐ5ɍÕÉ
3. Select Create Trivariate Histogram from the Mu ÁÂÕÄ$Å$ƏÕÅ$ÂÇ
sub-menu and specify the three source images in
the pop-up menu dialog.
The trivariate histogram will be calculated and displayed in a new image
window. If the displayed intensity ranges require adjustment, repeat steps 2-3
above using different survey technique options.

Figure 2-6 Trivariate histogram (with annotations) of the source images shown in Figure 2-5.

Multivariate Histogram Analysis User’s Guide Rev 1 2-7

 Extracting a single phase image from a histogram plot

The computed trivariate histogram is shown in Figure 2-6. The convention

followed when plotting the histogram is that the fractional contribution to a
pixel in the histogram is related to its perpendicular distance from the side
opposite the vertex corresponding to the species of interest. That is, clusters
with high fractional concentrations of a particular element will be situated
closer to the corresponding vertex. Hence in Figure 2-6, the fractional content
of oxygen with respect to the sum of the three constituents of all pixels along
the white line is related to the ratio of the length of the blue arrow to the red
arrow. Note, however, that as scaling is applied to the images before
computing the fractional contributions for reasons of display clarity, then this
fraction is only qualitative.
It can be seen that a number of clusters are present in the generated trivariate
histogram, indicating a number of well-defined chemical phases are present.
Extraction of the spatial distribution maps corresponding to these phases is
described in the two following sections.

Figure 2-7 Extracting a phase-specific map (right) from a bivariate histogram (left).

2.4 Extracting a single phase image from a

histogram plot

Once a multivariate histogram has been calculated, the next operation one
would most likely wish to perform is to extract the corresponding spatial
representations of selected histogram regions. This is can be achieved by
selecting a portion of the histogram using an ROI tool, and selecting the
EXTRACT SINGLE PHASE IMAGE sub-menu item. A trace-back algorithm then
extracts the phase-specific map corresponding to the selected histogram region
and displays it in a new image window. The phase-specific map is displayed as
a binary image, with pixels associated with the selected histogram region set to
unity and all other pixels set to zero. Note that to perform this operation, all
source images must be loaded into Digital Micrograph; if not, a suitable error
message will be posted. Additionally, no functions or transforms should have
been carried out on the original source images since histogram calculation (for

2-8 Multivariate Histogram Analysis User’s Guide Rev 1

 Extracting a single phase image from a histogram plot

example, a mathematical operation via the ÑbÒ*ÓÔ(Õ&Ö×"ØÙ(ÚÜÛÞÝOß&à5áãâ äß&Ý ), else the

trace-back algorithm employed will produce erroneous results.
An example of extracting a single phase-specific map from the bivariate
histogram generated in the previous section is shown in Figure 2-7 above. The
procedure for extracting a single phase-map is as follows;

To extract a single phase map from a multivariate histogram plot

1. Choose any ROI tool from the ROI Tools floating window, and
highlight the region of interest on the histogram plot.
2. Select the Extract Single Phase Image item in the Multivariate
Histogram Analysis sub-menu.
3. The corresponding phase image is then extracted and displayed in a
new image-window
As shown in Figure 2-7, once the phase map has been extracted the ROI is
displayed as a continuous line, indicating it has been made non-deletable (that
is, it is not deleted when another ROI is drawn, though you may delete it
explicitly), and labeled with a number corresponding to the title of the extracted
phase map. ROI placement is a somewhat arbitrary affair, and as a rule it is
best to select well-defined clusters using the closed loop tool. It can be seen in
the bivariate histogram example that the selected cluster of intensity is
associated with the carbon rich regions in the carbon map; this is consistent
with the cluster having high carbon and low Au jump-ratio values in the
histogram plot. Any number of phase maps may be extracted from a single
histogram by repeating the procedure outlined above.

Figure 2-8 Extracting a phase-specific map (right) from a trivariate histogram.

Similarly, a single phase may be extracted from a trivariate histogram following

the same procedure, as illustrated in Figure 2-8. In this example, the selected
phase is positioned close to the titanium vertex, suggesting that the phase is
titanium rich. However, a comparison of the extracted phase-specific map and
comparison with the titanium source image reveals that the phase-map contains

Multivariate Histogram Analysis User’s Guide Rev 1 2-9

 Extracting a composite phase image

regions of both high and low titanium content. The reason for this somewhat
ambiguous result is that the selected cluster of intensity corresponds to regions
of high titanium content, but also to other regions of low oxygen and aluminum
signal. As the plot displays intensities relative to the other constituents, the two
phases are combined in the same cluster of intensity. This example is a good
illustration of the qualitative nature of the trivariate histogram method
implemented, where the normalization procedure employed gives rise to well-
defined and well-separated clusters of intensity, useful when extracting phase
specific maps, but at the expense of quantitative information. As a result, a
little caution and thought should be applied when interpreting trivariate
histogram plots quantitatively.

Figure 2-9 Generating a composite phase image (bottom) from a bivariate histogram (left).
The color-encoded ROI map is shown right with ICP palette overlaid.

ROI selected
cluster

EXTRACT
PHASE

2.5 Extracting a composite phase image

In addition to extracting single phase-specific images from histogram plots, you

may also generate composite phase images built up from multiple phase-
specific maps. Composite phase maps are color (or grayscale) encoded for
clarity, using the color table specified by the selected åQæçè(æé2ê*ë(ìíîïé2ì5ð"ïè

2-10 Multivariate Histogram Analysis User’s Guide Rev 1

 Extracting a composite phase image

ñQòó&òôÜõö÷Vó&ø option in the ù0ú*û2ü òý5ô-öþmÿbø ü
 dialog. Selecting Qü ô-ö
ü
ñQòþ  ò û2ú*ü ø  ö û ø þöý5ø from the  ó ü(ú öô ú ö ü ø ù0ú*û2ü òý5ô-öþ$öó û2ú*û sub-
 øVô òô-ô-øVó&ö
menu initiates the ü ü(ú  øöô ü(ú*ü(ú ò ú $ý floating palette, as shown in
Figure 2-9. To extract a phase map from the multivariate histogram when in
this mode, place an ROI on the region of interest and select Qü
ô-
ö ü  ö û ø on
the palette. The corresponding phase map will then be displayed, color
encoded, in a new image-window. In addition, an ROI Map will also be
displayed, showing the selected histogram region encoded with the same color
or grayscale (see Figure 2-9). The composite phase map is built up by
extracting successive phase images from different regions of the histogram,
with each phase image being displayed in a different color. Note when in this
mode, the selected ROI region is removed from the source histogram display
after each successive calculation, preventing a histogram region contributing
twice to the composite phase map through overlapping ROI’s. If a mistake is

òšÿ ü ø
made,

or alternatively if you wish to alter your choice of region, select the
 button to undo the previous phase extraction. To complete the
analysis, select 

ú ú*û
on the palette to end the mode and restore the histogram
to its original state.

Figure 2-10 The composite phase image (right), with corresponding color encoded ROI map
(left).

An example of a composite phase map with the corresponding ROI Map is

shown in Figure 2-10. In the example shown, it can be seen that the gold and
carbon abundant phases correspond to the intensity clusters shown as green and
blue respectively in the composite ROI map, while the intermediate branch
joining the two discrete clusters corresponds to regions of overlap between the
edges of the gold particles and the surrounding carbon film as a consequence of
the particle’s spherical form. In this example, the two chemical distribution
maps have been reduced to a single image, with phases represented by three
distinct colors.
In summary, the procedure for calculating a composite phase image is as
follows:

Multivariate Histogram Analysis User’s Guide Rev 1 2-11

 Extracting a composite phase image

To extract a composite phase map from a multivariate histogram plot

1. Specify the desired color table in the !#"%$#&('*)+-,/.10&(+2"%0$%!#3#4
5
063+ pull down menu in the 78'*&()#9!4/0";:+2)<=$?> dialog.

2. With the multivariate histogram of interest front -most, select @#A!)4/0B2)

!#"%$#&('*)+-,/.10&(+ in the CD<=3)'*E04/'*0)+-78'*&()#9!4/0"GFIH=03J&('*& sub-
menu to initiate the KLH=)+24/B2#4/4/+230)'*E+-,/04/)'*)'*#H='*H=9 floating palette.
3. Select the histogram region of interest using an ROI from the ROI
TOOLS floating palette, and extract the corresponding phase (and ROI)
map by selecting EXTRACT PHASE on the floating palette.
4. Repeat step 3 to add additional phase maps. If you wish to undo one or
more steps, click on MNH=O=P:)+2$ .
5. When complete, select the Q'*H='*&(. button on the floating dialog to leave
the analysis mode and restore the original histogram.
It should be noted that each composite phase map color table has a maximum
number of available colors; when this limit is reached, an appropriate message
will be posted and no additional phase maps may be added to the composite
map.

Figure 2-11 Composite phase image (right) with corresponding trivariate ROI map (left).

Composite phase maps may be extracted from trivariate histograms in an

identical way. This is illustrated in Figure 2-11, where a composite phase map
and corresponding ROI map have been calculated from the trivariate histogram
generated in Section 2.3. As shown, the numerous clusters of intensity within
the trivariate histogram correspond to distinct, well-defined phases as shown in
the composite phase image. In this way, the information within the three
elemental maps has been reduced to a single image that suggests that a small
number of distinct phases exist in the dataset. These phase maps may be used
to extract phase-specific information from the source images or dataset;
functions incorporated to help do this are discussed in the followi ng two
sections.

2-12 Multivariate Histogram Analysis User’s Guide Rev 1

 Extracting a single phase image from a composite map

2.6 Extracting a single phase image from a

composite map

After generating a composite phase image, it is often desirable to extract one or

more of the constituent phases as a single image. The extracted single phase-
specific map could be used, for example, as a spatially specific mask to
selectively integrate regions in one or more elemental maps for performing
spatially specific atomic-ratio calculations or, alternatively, to perform phase
specific spectroscopy on a spectrum-image dataset as described in the following
section.

Figure 2-12 Single phase map extraction from a composite phase image (left) using the
EXTRACT SINGLE PHASE MAP FROM COMPOSITE option.

To extract a single phase-map from a composite, select the RS2T
UV

W(TYX[Z\]^_
`a
V
bc_dV
egfUhiGj2hiYe
hbcZT
_ option from the dk^T
Zl
V
UZV
T
_mnZbcT
h]UV
i
o
\V
^p
bcZb sub-menu with the composite phase map front-most. A dialog will
request you to specify which number phase you want to extract, as shown in
Figure 2-12. Each phase in the composite image has a value corresponding to
the sequence in which it was added (the first phase selected has a value of 1, the
second phase a value of 2, and so on). By placing the cursor on the desired
phase within the composite, the associated value can be viewed in the qiYV
]_
X[T
V
T
kb floating window. After specifying the desired phase for extraction, the
single phase will be displayed in a new image window. In summary;

To extract a single phase image from a composite phase map

1. With the multivariate histogram front-most, select r#s!tu/vw2tyxz*{=|!}~
/€
v(~-‚DvƒN„
u/…#†;‡!…#†%ƒ…#(z*t~ in the ‚Dˆ=}tz*‰vu/z*vt~-Š8z*(t…#|!u/v†
‹
{=v}Œ(z* sub-menu and enter the number corresponding to the phase
you wish to extract.

Multivariate Histogram Analysis User’s Guide Rev 1 2-13

 Performing phase-specific spectroscopy on a spectrum-image

2. The extracted phase map will be displayed in a new image-window.

It should be noted that the a binary phase map is extracted, with the value of the
pixels associated with the extracted phase set at unity (and zero for all other
pixels), regardless of its (non-zero) value in the composite phase image.

2.7 Performing phase-specific spectroscopy on a

spectrum-image

A phase-specific map provides an effective filter (or mask) for performing

phase-specific spectroscopy on a spectrum-image dataset. To carry out this
operation, generate a single phase map by either directly from a multivariate
histogram plot as described in Section 4.4 or, alternatively, by extracting a
single plane from a composite phase map as shown in the previous section.
Selecting Ž2
‘
’(Y“”#‘
•c–—™˜[š
–›’(œœ’N˜[š
–›’(
žŸ ¡ŸG˜¢ž•cœ£¤-¥‘
•c¦ in the
¥ž§
Ϭ
‘
œ‘

–©nœ•c
¡¤‘
Ÿª£‘
§«
•cœ• sub-menu initiates a dialog asking you to
specify the spectrum-image and phase-map to be used for this operation, as
shown in Figure 2-13.

Figure 2-13 Initiating phase-specific spectroscopy on an EELS spectrum-image (left) using a

phase-image (right).

It should be noted that the spectrum image and phase map should have
identically sized spatial dimensions to perform this operation; if there is an
incompatibility, an appropriate error message will be given. Once the
appropriate information is entered, the procedure checks the images are
compatible for the operation before multiplying each image plane in the
spectrum-image by the binary phase image. The filtered spectrum-image is then
summed in the spatial directions, and the resulting phase-specific spectrum is
plotted in a new image-window as shown in Figure 2-14.

2-14 Multivariate Histogram Analysis User’s Guide Rev 1

 Advanced notes

Figure 2-14 Phase specific spectrum generated as shown in the previous figure.

To extract a phase-specific spectrum from a spectrum-image using a phase-

map
1. Select ¬#­!®¯/°±2®y²/³1°´(µ/¶™·¸µ2±2¹*º
¹*±-·¸µ2±2®¯/»=¼½º
¯/¾#¼;·!¿/»=´(¹*À=ÁÃÂD°´(Ä in
the ÂD»=Å®¹*Æ°¯/¹*°®µ-Ç8¹*´(®¾#Á!¯/°¼GÈIÀ=°ÅÉ´(¹*´ sub-menu, and specify the
spectrum-image and (single) phase image in the appropriate fields.
2. The phase-specific spectrum will be computed and displayed in a new
image-window.

2.8 Advanced notes

The default preference settings for the Multivariate Histogram procedure are
contained in the global information tags. In addition to the parameters
accessible via the ÊnËÌcÍ
ÎÏÐÑ
ÒGÓ[ԛÍ
ÕÖ× dialog, additional items such as font size
and style for the trivariate histogram and histogram display sizes are contained
here. These values may be viewed and edited via the Global Info dialog box,
initiated by selecting the Ø#ÙÎڛÑ
Ù1ÛÜÝÎ#× menu item in the Þß àLá menu. The
multivariate histogram preferences are found in â!Ñ
ÏÌ under ãÕÙÍ
Ëä
Ñ
ÐËÑ
Í
Ô
ÊnËÌcÍ
ÎÏÐÑ
񌆄
Ùæ
ÌcËÌ , as shown in Figure 2-15.

Multivariate Histogram Analysis User’s Guide Rev 1 2-15

 Advanced notes

Figure 2-15 The Multivariate Histogram Analysis preference tags.

To alter the value of a preference tag, simply highlight the item, click çèéê and
enter the new preference. To revert a preference to its default (factory) setting,
delete the entry from the ëìí›îí›ìí›ïð(í›ñ group by highlighting it and selecting
ò
í›óYôõ
í , and the default entry will be generated next time the histogram routine
requires it.

2-16 Multivariate Histogram Analysis User’s Guide Rev 1

 Index

B I
Bivariate histogram Inter Correlative Partitioning, 2-8
creating, 2-3
example of, 2-5 M
extracting phase images from, 2-8
Multivariate Histogram Analysis
overview, 1-1
C
setting up, 2-1
Composite phase maps
creating, 2-10 P
extracting a single image from, 2-13
Phase extraction, 2-8
Phase-specific spectroscopy, 2-14
E
Extraction T
of single map from composite, 2-13
Tags, 2-15
Trace-back, 2-8
H
Trivariate histogram
Histogram creating, 2-5
color, 2-2 example of, 2-7
intensity range, 2-2 extracting phase images from, 2-9
resolution, 2-2

Multivariate Histogram Analysis User’s Guide Rev 1 I-1