Knowledgebase: Application Notes
Resolution and Accuracy
Posted by Elisha Byrne, Last modified by Micah Simonsen on 13 October 2016 01:05 PM

System Specs

Many things in our test set-up can affect our noise and resolution, but with a good setup, the following rules of thumb can be applied to estimate what type of noise and resolution you can expect out of your test.  This is what we base most of our system specs off of, although with excellent test setups, we can get much better resolution than spec'ed.

Spatial Resolution Rule-of-Thumb:
The spatial resolution will determined, in different ways, by both the subset and step.  Each data point in the output represents information that is averaged over a box that is subset size x subset size (i.e., 29 x 29 pixels). So this could be considered as the spatial resolution.
Each data point is separated from the next by the step size. However, if the step is less than the subset size, the two points are not completely independent, and if the step is much smaller (i.e., subset 29, step 1), the points are not very independent at all.  Also, for the normal setting (Subset weights... Gaussian), the subset is center weighted, so it actually represents a somewhat smaller area than the subset size.

Displacement Accuracy Rule-of-Thumb:
In typical setups we see an in-plane displacement accuracy of 1/50,000 of the field of view.  So if we are imaging a 1 meter area, we can detect a 20 micron motion.  For out-of-plane, it's about half that 1/25,000 of the field of view.  For higher resolution cameras and/or particularly good experimental setups (good pattern, lighting, etc.), the displacement accuracy can be much better than 1/50,000.  However, most systems are spec'ed out using the more conservative 1/50,000 rule.

Strain Resolution, Rule of Thumb:
Our strain resolution for a typical setup is within 100 microstrain.  Again, with a very good test setup, we can get much better.


Measuring and Calculating Accuracy and Resolution for Specific Test Setups

Noise levels of a test are very specific to the test setup.  Many things can increase noise:

  • Focus
  • Lighting
  • Glare
  • F-stop
  • Stereo-Angle/Lens Selection
  • Subset size
  • Speckle Pattern Quality

For more on how the above items affect noise and resolution, see here:

Because of the variability in noise level from test-to-test, it is often beneficial to examine the noise level of your specific test.  There are a few different ways to do this. 

Displacement Accuracy, Calculated (sigma):
The accuracy can be determined by clicking the "Compute confidence intervals" box in the Postprocessing tab of the Analysis dialog. You will obtain a Sigma_X, Sigma_Y, Sigma_Z. These are the 1-standard-deviation (67%) confidence intervals for that point. You may double this to get 95% confidence intervals. This is an accurate noise estimate, although it ignores bias (for example, if you had a bad calibration, it would not be reflected here).

Displacement Accuracy, Measured (u, v, w fields):
Since the sigma values don't reflect any bias in the system, a more conservative estimate of the displacement fields would be to take a few static images, and examine the u,v,w fields for the static images of your sample.  Specifically, the average and standard-deviation (right click on contour plot>statistics), are of interest here.  This is still a conservative estimate because no rigid motion has occurred.  So for a very realistic measurement, if you can rigidly move the sample (about the amount that you expect it to move during the test).  If you can rigidly move the sample with no strain, then we can look at the displacement fields for an extremely accurate measurement of noise.  Once the translation images are ran in Vic-3D, remove the rigid motion (data>post processing>remove rigid motion).  This sets the u,v,w fields back to 0.  So the remaining local motion (now apparent in the u,v,w fields) is our noise.  Right click on the contour plot to view the standard deviation for u,v,w once the rigid motion is removed.  This is the most conservative measurement for your displacement noise.

Strain Resolution, Measured:
Using the same rigid motion test as for the "Displacement Accuracy, Measured," now look at your strains.  This is your strain noise.  You can examine both the mean and standard deviation by right clicking on the contour plot and clicking statistics.  Note that you actually don't NEED to remove rigid motion in the software, as strains are unaffected by rigid motion.

Virtual Strain Gauge Size
When selecting the filter size for strain, keep in mind that this is in terms of data points, which are separated by the step size. So if your filter size is 15, and your step is 5, the total smoothing area is 15*5 = 75 pixels. If you reduce the step to 1, and use a 15 filter size, you will only be smoothing by 15*1 = 15 pixels, so the strain will be noisier.  The strain filter is also center weighted.

(11 vote(s))
Not helpful

Comments (0)
Post a new comment
Full Name: