i've been playing around with a few different piv setups (different cameras, etc), and i was wondering if there's any quantitative metric for how 'good' or accurate a given PIV analysis is in pivlab to compare my various setups. is there a definitive metric, or do people have suggestions?
You can judge the precision and accuracy of the software itself by analysing artificial images with known properties (see my chapter on PIV in my thesis).
If you want to judge the quality of your setup, I think the best way is to calculate the signal to noise of the cross-correlation. PIVlab doesn't save the result of the cross-correlation, so adding this option is a bit more work. I am however considering to add this.
As far as I remember, there are two ways to calculate the s2n: Ratio of first peak height to second peak height, and something else that I forgot. The details are given in Raffel et al. 'PIV a practical guide'.
Thanks for the reply! PIV a practical guide has been my bible for the past few weeks, I'll check it out to find s2n calculations. Will I need to dive into the PIVlab code to grab the results of the cross-correlation? or is there an easy way you know of doing this?
Just following up after having spent some time on this:
I can find the cross-correlation of two images using matlab, but i'm not sure how to recreate the right interrogation window sizes. is getting the results of the cross-correlation from PIVlab a straightforward task, or is this a relatively laborious endeavor? I am just trying to gauge whether or not it is worth investing more time in.
The result of the cross correlation is in PIVlab_GUI.m, the variable is called result_conv. You could edit the code and save this variable for each interrogation window.
This shouldn't be too much work for a quick and dirty solution, but it is quite some work to add it properly to the GUI.
Ah, sorry, you have to open piv_fftmulti.m...
You should look at the result_conv in the function 'subpixgauss', there you'll have all the correlation results of all the passes for all interrogation areas.
thanks for all the help, this worked out! my last question is, to present the s2n of a given cross-correlation, is it best to average the s2n over all interrogation areas of the same size? basically i have all the s2n for all of my interrogation areas for all passes, but i'm not sure how to turn all of this into a 'metric' for the analysis overall. i couldn't find any info on this in raffel et al.
What method did you choose for calculating s2n? Peak-to-second-peak ratio?
I could imagine that there are many ways to calculate one single 'analysis quality' parameter out of the s2n numbers.
Calculating the mean +stdev is probably the simplest, and maybe it is enough for your purpose.