Edge preservation near triple junctions during orientation averaging of EBSP data
Introduction
For electron backscatter pattern (EBSP) investigations, measurement many times within a region of the same orientation results in a distribution of orientations, that can be thought of as adding an “orientation noise” to the data [1], [2], [3], limiting the angular resolution of the technique. In order to improve the angular resolution, the use of a modified Kuwahara filter on EBSP data has been suggested [4]. The requirement for such noise filtering is particularly relevant for investigations of deformation microstructures, as during deformation new dislocation boundaries are continually introduced, contributing to a class of low misorientation angle boundaries, referred to elsewhere as incidental dislocation boundaries (IDBs) [5]. Even at high strains the average IDB misorientation remains low (e.g. 2.7° for Ni (99.99%) rolled to 90% reduction [6]), and these boundaries can comprise nearly half of the total dislocation boundary area per volume.
For EBSP maps, microstructures must be probed by first defining an imaging misorientation level θmis, where misorientations of >θmis are regarded as corresponding to real features. There are, therefore, two aspects to consider when evaluating the angular resolution: (i) to reveal features with misorientations below a certain value, and (ii) to demonstrate that no other features exist with misorientations below the lowest real misorientation. Given these two aspects, the primary objective in orientation filtering should be to allow the lowest possible value of θmis, providing the most information for a given map. It is important, however, not only to reduce the average noise level, but also to limit the number of artifacts introduced during filtering, as these may treated as real features by an automated analysis of the data.
In this work a feature-based methodology is first presented for relating the use of Kuwahara-based orientation filters to the noise level in the data, making use of simulated EBSP maps of single crystals and dislocation cell structures. The results are then used to suggest a further improvement to the filter.
Section snippets
The modified Kuwahara filter
Application of the modified Kuwahara filter is performed as follows (a more detailed description is given in [4]). At each point in the EBSP map a square filter grid, centered on the pixel of interest, of size (2N−1) pixels square is considered. For all N×N sub-grids fully contained within the filter grid and that extend to the filter grid edge,1 the mean and variance of the orientations are calculated. The point
Simulated EBSP maps of single crystal material
The orientation noise resulting from EBSP measurement within regions of perfect single crystal material was simulated using maps of size 100 × 100 pixels. Noise was added to give different values of 〈θi,j〉pre, the mean misorientation between adjacent map points before filtering. For each noise level, five simulations were carried out, each using a different seed for the random number generator used to simulate the orientation noise. The noise was simulated such that the distribution functions of
Discussion
The simulations show that a feature-based methodology can be useful in suggesting guidelines for application of Kuwahara-based orientation filters. It should be remembered though that 〈θi,j〉pre can only be measured directly on undeformed crystals. Values for a deformed sample will be higher, as a result of lower pattern qualities and greater likelihood of pattern overlap, and may also vary between boundary and interior regions. Further work to obtain a better understanding of the extent and
Acknowledgments
Discussions with Drs. D. Juul Jensen and J. Bowen are gratefully acknowledged. This work was supported by the National Natural Science Foundation of China under contract no. 59825110.
References (9)
- et al.
Acta Mater.
(1997) - et al.
Acta Mater.
(2000) - et al.
Acta Mater.
(2001) J. Microsc.
(1999)