SOME IDEAS ON MEASURING ACCURACY WITH ELECTRON MICROGRAPHY
by
Sanjib K. Ghosh
Department of Photogrammetry
Laval University
Québec, G1K 7P4
Canada
ABSTRACT
Accuracy of measurements with electron micrography seems to depend on
factors like Distortions, Scale repeatability, Intersection geometry, Instru-
mentation, Magnification, etc. Some ideas on these obtained from recent
research investigations are presented with regard to one SEM system.
RESUME
La précision des mesurages par les micrographies électroniques, à ce
qu'il paraît, dépend des choses comme les distortions, la capacité pour répé-
ter l'échelle, la géométrie de l'intersection, l'instrumentation, le gros-
sissement, etc. Quelques idées sur celles-ci, qui ont été obtenues par des
recherches récentes, sont présentées, Ce sont aux égards d'un système SEM.
INTRODUCTION
In extracting quantitative data with electron micrography (EM), as with
any other system, one is concerned about the assessment of their quality.
This assessment, in practice, is often made by considering certain properties
related (1) to the statistical behavior of the observed and extracted data;
and (2) to the nature of the measuring system they refer to. While the for-
mer bears upon certain theoretical concepts, the latter pertains to the
physical reality. In dealing with the mensural EM data they both are equally
important and need some elaboration. Furthermore, such quality considera-
tions are necessary both in designing and in evaluating the measuring
process.
The quality or the degree of perfection in a measurement is expressed
in terms of accuracy (defined in the Webster's dictionary as: degree of
conformity of a measure to a standard or a true value). The accuracy requi-
rements for a measuring process are regulated partly by experience, partly
by conference (i.e., discussions and exchange of ideas with various persons
from various fields associated in the system) and partly by analyses. Howe-
ver, the observations, i.e., the outcome of a measuring process, are assessed
in practice in terms of "accuracy" and "precision". These two terms are
confusing to the uninitiated. It will therefore be pertinent to explain them.
Without going into deeper theoretical statistical connotations, it can
be said that "accuracy" indicates the degree of closeness of the observations
to the "true" value or the "input". "Precision", on the other hand, indica-
tes the degree of closeness of the observations to their average value, i.e.,
the degree of agreement or uniformity among a set of observations of the
same random variable.
The relationship of the precision to the corresponding mean (or average
value) is almost parallel to that of the accuracy to the true value. That
218