Extracting morphological changes in nanocrystals using in situ liquid cell microscopy
Two-point statistics and Further Work
As previously noted, a two-phase system was a natural choice in our case because our physical domain consists only of liquid solution and solid nanoparticles. This is why binarizing the image was so important. Now that we have a binarized image, we can proceed with 2-point statistics. For the most part, the PyMKS toolbox makes this process slightly trivial since the calculations are so well-documented and the scripts provided. Below you can see the one of the auto-correlations for our system.
Since the system is two-phase, the correlations are linearly dependent so we do not need to calculate the other auto-correlation or the cross-correlation; it would be redundant. Visual inspection of the figure can yield information on the volume fraction, radial symmetry, and characteristic length/particle separation of the system. However, the above figure is for periodic boundary conditions. This is in fact not the ideal parameter for our case since the microstructure is not strictly periodic in nature. Therefore the two-point statistics must be re-calculated and then compared to the periodic case.
Now that two-point statistics have been achieved, we must move forward to PCA. Also, we need to polish our process so that we have more refined binarization and fully automated frame extraction from the raw data sets. Currently we must manually select fully refreshed frames, but we are working on an automated routine to extract, crop, denoise, filter, and finally binarize the complete frames so that two-point statistics may be implemented. Our next blog will update our progress on that, as well as PCA. Please feel free to comment!