Image analysis for Materials Science

The combination of new technologies, new software approaches, and expansion

of analytical techniques to include more automated stereology is expanding the
ability of materials scientists to define microstructure and determine its effect on
function.
Recent advances in both computer and camera electronics combined with older
but less accessible stereological analyses are opening new venues for
microstrucural analysis. The most important areas of impact for these advances
are a better understanding of the structure/function relationship and resulting
process control. This article reviews some of the current trends in technology and
cites several applications of stereology.

Windows-based technologies
For decades, microstructure has been analyzed quantitatively through either
laborious manual methods or fully integrated image analysis systems. However,
advances in desktop computers and Windows-based software are opening new
opportunities for customized systems. These are built from components and
configured by system integrators or value-added resalers who carry a range of
cameras and monitors, digitizing boards, and software solutions. The new
systems can be built onto or networked with existing computers, saving money
and minimizing training.
These advances also enable software developers to offer greater functionality at
lower cost. The Windows interface provides a venue for control of peripherals
such as scanning stages (x, y positioning and scanning patterns) and focus (z),
shutters, and filterwheets, considerably reducing the time necessary for
generating computer code. Microscope companies such as Nikon and Zeiss have
been taking advantage of this Windows feature for the past several years.
Recently, off-the-shelf software companies such as Optimas and Media
Cybernetics (Image Pro Plus, Fig. 1) have built new control modules into their
packages, specifically for the stages and other peripherals offered through Prior
Scientific and Ludl Electronic Products.
Fig. 1 (right) - Image Pro Plus has
built new control modules into their
packages, specifically for the stages.

Windows-based programming also expands report-writing capabilities, readily

integrating text, images, graphics, and data as well as extending control from
desktop computers to the process line. Analysis, reporting, and process control
from one central location or via a network is now a reality.

New camera technologies

Development of this new generation of custom integrated systems has also been
spurred by advances in camera technology, Cameras such as the Leaf Lumina,
and combined camera/electronic systems such as those available from Dage-
MTI, offer greater resolution at lower prices. In addition, advances in color
technology enable use of software packages offering true color segmentation in
addition to traditional thresholding by gray-level.
The next challenge for designers will be to provide increased computer
processing capability to match the larger image formats from these cameras.
Current minimum standards for "high resolution " image are 512 x 512 pixels in
black-and-white mode, requiring that the computer system handle a quarter of a
megabyte of data per image. However, higher XY resolution capability is
pushing that requirement rapidly into the one to four megabyte range. Adding
color multiplies that requirement by nearly an order of magnitude, stressing all
but the workstation and mainframe platforms. Materials scientists new to image
analysis or considering expansion of existing systems should carefully assess
their needs and the ability of the computer portion of their system to handle the
output and processing requirements of these new technologies.

Innovative approaches
Windows technology is not the only advance on the programming front. The new
direction in "shrink-wrapped" software is toward application-specific packages.
In the most progressive systems, such as those offered through Media
Cybernetics recently released Mat Pro Materials Science Module, the interface
uses conventional materials science terminology and follows standard protocols.
This approach simplifies analysis by isolating key steps.
Both the more expensive fully integrated systems and the off-the-shelf software
provide the ability to construct macros, lists of instructions that automate routine
tasks, In most cases, no programming or even typing skills are required. Rather, a
mouse is pointed at the appropriate function and simply clicked to add that step
to the instruction list.
Although some systems still rely on gray-level thresholding, many manufacturers
are expanding their programs to include real color, an advantage in many
semiconductor applications, polarized light analyses, and certain acid-etched
metallic surface studies. Other systems, such as Context Vision, are even
segmenting based on texture.
Grain-boundary completion has always been a challenge for more automated
systems, but advanced software such as Visilog (Noesis) and software/hardware
solutions such as Imagist (Princeton Gamma Tech) solve the problem. For
example, Noesis uses a novel combination of watershed and reconstruction
algorithms to reconstruct the missing grain boundary, while PGT has
incorporated a version of artificial intelligence to make local comparisons
between hypothetical images to reconstruct the missing grain boundary.

Changing microscopy
In a radical departure from conventional microscope design, prototypes now
available through General Scanning Inc. (GSI) and materials sciences
instruments from Biomedical Photometrics Inc. are built specifically for
examining ultra-large fields of view and for high-throughput situations.
As shown below, the field of view (FOV) for a conventional microscope may be
calculated by dividing the field number engraved on the eyepiece (usually
ranging from 18 mm to about 26 mm in newer metallographs) by the total
magnification of the objective and any intermediate optics.

FOV = field number /( Mobj * Mint)

For example, for a IOX objective viewed with an eyepiece having a field number
26, the field of view is 2.6 mm. While an objective of this type will have a
reasonable working distance (2.3 to about 7 mm), its resolution would be limited
to somewhere between 0.75 and 1.4 um.
In comparison, specifications on the new large-format microscopes are
impressive. By clever application of electronics, they require only one objective
to scan magnifications from 1X (5 x 5 gm) to 10OX (500 x 500 mm or IX FOV)
on a 1024 x 1024 display. An electronic pan mode enables the viewer to see any
portion of the field on demand. Spectrally, they detect color from the near ultra-
violet through the visible spectrum (365 to 680 nm). The GSI system, for
example, provides a numerical aperture of 1.00 (the theoretical limit for non-
immersed optics), offering submicron resolution (Rxy = 0.5 um) over a 5 mm
field of view, with a 3 to 4 mm working distance, a remarkable combination of
specifications.
For those applications in which information is obscured by glare or haze (rough
surfaces or polyurethane foams), both microscopes can operate in confocal mode,
cleanly imaging these previously inaccessible structures.

Structure-function relationships
The greatest impact of the new technologies is on a stronger understanding of the
structure-function relationship in materials, and on the ability to translate data
gathered from microstructures into better process control. Traditionally, much
image analysis has been based on the geometric information (area, perimeter,
projection, etc.) derived from the feature-specific analyses in morphometry.
While the histograms of population distributions generated by this approach can
be valuable, all too often they fall short of truly describing the bulk properties
that influence mechanical strength, thermal sensitivity, and other properties.
Stereology provides the necessary field-oriented analyses, but it requires of a
large number of fields to be statistically valid. Currently, relatively few
automated image analysis systems include any stereology, necessitating hours of
tedious manual work. To further complicate matters, the mathematical proofs
behind this method are extremely challenging. Texts written on the subject are
aimed at communication within that field of mathematics rather than for the lay
user, making entry daunting, and masking relevance to everyday applications.
Ironically, the actual measurements needed for stereology are extremely simple
and available in the digitized images common to all image analysis packages, and
the actual math is trivial and easily inserted. The result is a body of parameters
that provide strong relationships between structure and function. They can be
readily translated into process control of factors such as time at temperature,
moisture content, and concentration of constituent or catalyst.
Fig. 2 (right) - A stereological
application using Optimas
software to determine grain
size. In the example shown
here, users may choose
between cycloid, grid, or
diagonal line patterns.

Sizing operations
The ability to characterize powders, grain and cell sizes, and separated features
within a microstructure falls under the generic term "Sizing." Since micro-
features are often irregular in size, shape, and orientation, no single parameter
can be used to characterize their true form. However, for the population
distributions typical of morphometry, if the shapes of features are known, then
size distributions can be derived, and vice versa.
For regular polygonal grain structures, individual components can be assumed to
have a pseudo-spherical shape, and a simple intercept length and area
measurement on individual features, recorded from the plane section, can be used
to define the grain size distributions.
For example, to evaluate ferrite grain size in low-carbon steels, sizing can be
done by the linear intercept method. In this approach, a test pattern of lines of
known length is randomly laid over a plane section, and the individual intercept
lengths across features of interest are measured. The data is ranked, then grouped
into an arbitrarily determined number of size bins (such as 10 bins) of set size
increment (such as 2.5 um).
The image presented by the plane section will be that of grains having a
superimposed test line (pattern). The pattern intercepts the grains and produces
chords of varying lengths, with a maximum length equal to the diameter of the
sphere. Using the working relationship shown below, the number of intercepts
per unit test length (NL)can be used to find the number of grains of a specific
size per unit volume of the sample:
Where j=class, Nvj = number of grains/volume in that class size, nL= number of
intercepts/test line in that class, and delta=2.5 x 10-3mm, the intercept size range.

Intercept size Grain dia. No. of grains

Group #, Number,
range Dj, (X105) Percentage
j NLf/mm
Delta, um X10-3 mm per mm3, Nvj
1 0.0-2.5 6 2.5 68 5
2 2.5-5.0 17 5.0 177 15
3 5.0-7.5 24 7.5 426 35
4 7.5-10 19 10.0 259 21
5 10.0-12.5 13 12.5 146 12
6 12.5-15 8 15.0 70 6
7 15.0-17.5 5 17.5 24 2
8 17.5-20.0 4 20.0 19 1.5
9 20.0-22.5 3 22.5 14 1
10 22.5-25.0 2 25.0 21 1.5

For example, using the data from the table, terms for the fifth group give:

Nv5 = 2.04X105 [13/(10 - 1) - 8/(10+1)] = 146xlO5 grains/mm3

These results are based on only 101 intercept measurements, yet they are
statistically valid. In other words, accurate grain measurements may be achieved
without expensive equipment, by the use of an optical microscope, a simple
counting technique, and a ruler.