Data processing
If the data have been collected automatically as maps, the data will need to be corrected for non-indexed and mis-indexed points and is essential to eliminate errors and artifacts. The amount of data processing required will depend upon how well the raw data indexed and what the data will be used for (Bestmann and Prior, 2003; Prior et al., 2002; Toy et al., 2008). Once the raw EBSD data has been collected (Fig. 1c) a good way to start is to remove the ‘wildspikes’, single data pixels that are surrounded by 8 pixels of a different orientation. When the software removes each ‘wildspike’ it assigns the pixel an average orientation of the surrounding 8 neighbours (Fig. 1d). This is only appropriate in raw data sets with high indexing rates and low mis-indexing rates (Prior et al., 2009). The justification for doing this first is that the next step of data processing involves ‘growth’ and growing incorrect data points could introduce artifacts into the data.
It is common in geological raw data sets for indexing and mis-indexing rates to be heterogeneous whereby one grain is fully indexed and contains no errors yet the adjacent grain is only, say, 80% indexed and contains mis-indexed points. In these cases it is necessary to ‘grow’ the grains using the extrapolation routines in HKL’s Tango program. The Tango software allows non-indexed points to be filled with the average orientation of a specified number of neighbour pixels (where their orientation is the same within a defined tolerance). The software allows the number of neighbours to be altered to allow more or less growth to take place. Growth to completion on the basis of 8, 7, 6, neighbours does not create significant microstructural artifacts, nor does a single step at 5 neighbours (Prior et al., 2009). Figure 1e shows how the data looks after the wildspikes have been removed and growth to completion at 8 and 7 neighbours, whereas figure 1f shows the data after growth to completion at 6 and 5 neighbours plus one step at 4. Comparison of figure 1b and 1f shows that the boundary structure is still equivalent and no artifacts have been introduced. If you compare the grain that is inside the yellow circle between each of the maps, it shows how the ‘wildspikes’ (green coloured pixels) are apparent in figure 1c but removed by figure 1d and that non-indexed points (black pixels) are completely removed after the data processing is complete (Fig. 1f). After this type of data processing has been completed, continued growth based on fewer neighbours can introduce significant errors into the data set. Introduced errors can be identified and avoided by continuously comparing the processed data with the pattern quality map (which for the purpose can be assessed like a backscattered electron image). As soon as grains ‘grow’ so that the new positions of their grain boundaries cross those of the pattern-quality component map, then artifacts have been introduced and the data processing has gone too far. For more complicated ways to perform data processing see Prior et al., (2009).