46
S. P. Donegan and M. A. Groeber
Fig. 17 Original BSE tile (left) and corresponding binary segmented image (right)
Fig. 18 The mask of the identified component in each of the resampled datasets: EBSD (right),
BSE (middle), and DEFORM ® (right)
For the BSE montage, a 15 μm window was passed over the segmented highresolution montage and the average value of the binary segmentation was computed
within this window. This procedure yields an alpha area fraction at each 15 μm
pixel.
The resampling procedure conveniently brings all datasets onto the same geometric topology; however, they are still misaligned relative to one another. To register
the datasets together, the component was first identified in each of the modalities.
Figure 18 shows these component masks.
Since the goal is to determine correspondence between process zones and
microstructure, we do not wish to a priori assume any relationships that would bias
the registration. Instead, we identify the component geometry in each modality since
we expect it to be relatively invariant between each dataset. Thus, the component
geometry itself can be used as a registration datum. To obtain points for registration,
the exterior of the masks shown in Fig. 18 are regularly sampled. These sampled
points are shown in Fig. 19.
S. P. Donegan and M. A. Groeber
Fig. 17 Original BSE tile (left) and corresponding binary segmented image (right)
Fig. 18 The mask of the identified component in each of the resampled datasets: EBSD (right),
BSE (middle), and DEFORM ® (right)
For the BSE montage, a 15 μm window was passed over the segmented highresolution montage and the average value of the binary segmentation was computed
within this window. This procedure yields an alpha area fraction at each 15 μm
pixel.
The resampling procedure conveniently brings all datasets onto the same geometric topology; however, they are still misaligned relative to one another. To register
the datasets together, the component was first identified in each of the modalities.
Figure 18 shows these component masks.
Since the goal is to determine correspondence between process zones and
microstructure, we do not wish to a priori assume any relationships that would bias
the registration. Instead, we identify the component geometry in each modality since
we expect it to be relatively invariant between each dataset. Thus, the component
geometry itself can be used as a registration datum. To obtain points for registration,
the exterior of the masks shown in Fig. 18 are regularly sampled. These sampled
points are shown in Fig. 19.