...
Once the images were obtained, some data processing had to be performed in order to get the desired results and filter out erroneous data. Image pairs were processed with a subwindow size of 16pix x 16pix. This size was somewhat arbitrary, but was hoped to capture high image correlation between pairs. This choice of subwindow size resulted in a map of mean displacements in each subwindow. Using an image of a ruler to calibrate distance as well as knowing that the camera image rate was 30 Hz, these displacements were then converted to velocities by the equation:
The 
The U and V displacement matrices were also run through a simple band pass filter that set a tolerance for upper and lower bounds of displacement and cut out background noise. Histograms of U and V data for each temperature data set were examined and upper and lower bounds were chosen based on visual inspection. If a data point appeared outside of these bounds, the data point was set to the mean value at that data point.
Figure 6 - Band Pass Filter schematic and generic Matlab code used to process the data.
...
Figure 8 - Vector map of mean displacements at 50°C.
Figure 9 - Vector map of mean displacements at 70°C.
...