You are on page 1of 3

THE ART AND APPLICATION OF DIC

All about Speckles: Edge Sharpness


by Phillip Reu

Introduction
The size of the speckle has been specified and that the contrast should Properties of a speckle
be optimized – but what should the speckle look like? That is, do we 1. Size (pixels)
want a sharp edge or a softer edge on the speckle? As we discuss this 2. Contrast (grey levels)
topic it is important to appreciate that the real question is: What is the 3. Speckle edge sharpness (contrast
best looking speckle I can practically apply in my experimental context? gradient)
4. Speckle density (spatial distribution)

Gradient comparison between hard and soft edges


We are specifically discussing the edge profile of the
speckle in grey values. What I mean by this can be seen
in the side figure where both a ‘‘sharp’’ and a ‘‘soft’’
edged speckle are illustrated. A soft-edged speckle will
typically have a couple of pixels of transition between
the white and the black regions. A series of four speckle
patterns and line cuts are shown, the first is the hard
speckle edge, second is a soft focus image, and the final
two are created with post-processing using a Gaussian
filter.
Generally, a soft speckle edge is preferred1 . Why is this?
Comparison of edge sharpness.

The Problem with a Sharp-Edge


I suspect the problem with sharp-edged speckles is not
due to problems with the interpolant2 , but rather due to
the sampling of the speckle edge by the digital camera.
A good interpolant will always match the values at the
pixel locations, but the high frequency analog content
of the step-transition at the sharp edge will be aliased
by the detector. A sharp-edged speckle is like a square
wave signal that is sampled by the pixels. (I often
find this easier to think about in the time domain – a
square wave sampled by a data acquisition board.
Nevertheless, the principles are exactly the same!) If
you represent the square wave as a series of sine waves
added together3 , you see that there is a ‘‘fundamental A soft speckle edge showing some grey values in the transition
frequency’’ the width of the square wave, however to from white to black.
capture the sharp edges, there are a number of higher
1 Pan,
frequency sine waves, which are smaller than a pixel, B., Optics and Lasers in
needed to represent the sharp edge. This is illustrated Engineering, 2013. 51(10):
p. 1161–1167. (Note: This paper is mostly
concerned with noise rather than speckle
The Art and Application of Digital Image Correlation is written by Phillip L. Reu edge profiles.)
2
(Phillip.Reu.DIC@gmail.com). He received his PhD from the University of Wisconsin–Madison Or maybe more accurately, the
and is currently a Principal Member of Technical Staff at Sandia National Laboratory. He interaction of the interpolant with the
began working with digital image correlation in 2004 and is focused on understanding aliased data.
3
the influence of the unavoidable compromises made in field measurements on the final This is the same as doing a Fourier
DIC uncertainty. Sandia is a multiprogram laboratory operated by Sandia Corporation, a transform.
Lockheed Martin Company, for the United States Department of Energy’s National Nuclear
Security Administration under contract No. DE-AC04-94AL85000.

Experimental Techniques 39 (2015) 1–2 © 2015, Society for Experimental Mechanics 1


The Art and Application of DIC

4 in the side figure showing a sharp-edged speckle, along with the


See the aliasing discussion in a previous
article. fundamental and next two harmonics. Now you can see that the
fundamental of the square wave is not aliased4 , i.e. it spans 3 pixels,
however the higher frequency components are less than a pixel and
are aliased. The negative effects of this can be seen in the bias error
plot shown in the following figure. The higher frequencies (smaller
aliased speckles) cause larger bias errors decreasing the accuracy of
the correlation. This is analogous to adding noise to the speckle image,
and as we already know: Noise is bad.
Now the news isn’t all that bad in practice. It is extraordinarily
difficult to create a perfect square wave. In fact, you can see in my
previous example, that even the hard edged speckle has a pixel or
two of contrast at the transition. This is good, because many popular
speckling techniques tend to have harder-edged speckles, including;
spray paint, sharpie marking and stamping.
Illustation of a square wave (sharp speckle
edge) as seen by the pixels. Note that the
higher harmonics are smaller than a pixel and Obtaining a Soft Edge
are aliased. A number of approaches can help create a soft-edged speckle. By
printing your pattern on the sample using an inkjet printer, you have
complete control over the speckle size, shape and contrast. Stencils
also tend to work well, because there is usually a little over-spray that
will soften the edge.
There are two tricks you can use with the lens to add filtering that
will soften the speckles. First and easiest, is to provide a little defocus.
Unfortunately, this technique should not be used with stereo-DIC,
as it is impossible to get even defocus in both cameras due to the
perspective view of the stereo angle. Second, with some lenses near
their resolution limit, you can decrease the aperture to provide
filtering. Unfortunately, this method is harder to control, and may
Illustration of the bias error due to speckle decrease the contrast.
aliasing. Post filtering of the images will also soften the speckles and remove
noise from the image. Indeed, most programs offer some form of
image filtering in the analysis portion of the software and there is a
Exercise for the Reader
growing literature that seems to suggest that pre-filtering will improve
Setup a simple 2D translation the results.
experiment. Create a speckle pattern with
very sharp edges by printing a pattern on
a laser printer. Image the sample while Conclusions
translating it. Defocus the camera slightly
to give soft edges. Translate again while As a best practice, it is always better to optimize contrast, speckle
acquiring images. Compare the size, and distribution at the cost of losing the ideal speckle edge.
displacement noise between the two sets. This is why stencil patterns and sharpie markers are very popular
Now, take the original sharp edged speckle application methods. Often, when attempting to get a soft
speckles and filter them with a Gaussian
speckle edge with perfectly sized and distributed speckles, particularly
filter and run them again. What are the
results?
with spray paints, you end up with too much grey in between the
speckles, and lose contrast. Trying to balance all of these parameters
is difficult. Therefore, as a general rule, one can sacrifice the perfect
speckle edge to optimize the other three parameters. If needed,
the speckles can always be softened by the afore mentioned image
filtering methods. Optimize the contrast and speckle size first.
Next Time: Speckle density.

2 Experimental Techniques 39 (2015) 1–2 © 2015, Society for Experimental Mechanics


Copyright of Experimental Techniques is the property of Wiley-Blackwell and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder's express written permission. However, users may print, download, or email articles for
individual use.

You might also like