Prakt. Met. Sonderband 38 (2006) 95
LOM and 6. SUMMARY AND OUTLOOK
grains are
It. Using only light optical microscopy (LOM), the characterization of multiphase steels cannot
mination, be achieved sufficiently due to the fact that very fine metallographic constituents are
separate present, which are smaller than the optical resolution limit. This problem can be avoided by
fication is using scanning electron microscope (SEM) images of a higher resolution in addition to the
light optical images.
Common commercial pixel - based software programs for metallographic applications are
not able to analyse the phases in multiphase steels correctly. Problems occur due to the
usage of only the grey value in the observed image, during the classification process. If
there are small differences in grey values, the different metallographic constituents cannot
be separated from each other and a correct quantification is not possible using pixel -
based algorithms. Therefore, a new approach was developed for the analysis and
quantification of these steels.
Within this work, a method for automatic composing LOM images and SEM images into a
single feature image is described. A novel segment based approach for the analysis of
multiphase steel images is presented. Based on the composed feature image and its
gradient images, segments are calculated, which provide information about the shape of
ferrite and martensite. In addition, about 40 different features are provided by the
calculated segments in conjunction with the feature images. Thereby, contour, optical and
texture features as well as mean and variance values of the original and gradient images
and, in addition, neighbourhood relations, are derived. These features are used for the
classification of martensite and ferrite phases, using a Bayesian classificatory.
This segment based image analysing approach can be used for image analysis of various
materials. Pictures like EBSD, optical pictures from different etchings and SEM pictures
nase steel can be added to each other to provide image information from different receiving
and SEM technologies and make them useable for quantification.
tis about For future work even more features can be used for classification. A higher amount of
groundtruth — data will result in a better classification. Also more complex multiphase
Cs can be steels with a higher amount of metallographic constituents like TRIP steels could be
analysed and quantified by this method. In this way a new method is available meet
challenges in metallographic characterisation of complex materials
REFERENCES
[1] Ohm, J.-R.: Multimediakommunikation Skript SS 2002
[2] David G.Lowe: Distinctive Image Features from Scale-Invariant Keypoints.
Computer Science Department University of British Columbia Vancouver, B.C..
Canada, June 6, 2003
[3] llja N. Bronstein, Konstantin A. Semendjajew, Gerhard Musiol: Taschenbuch der
Mathematik, Verlag Harri Deutscher, 1995
[4] Dorin Comaniciu: Mean Shift: A Robust Approach Toward Feature Analysis Vol. 24,
5 May 2002
[5] Soille, Pierre: Morphologische Bildverarbeitung. Springer Verlag Berlin, Heidelberg
1998
[6] K -F. Kraiss: Mensch-Maschine-Systeme Skript SS 2004