Fundamentals of Digital Image Correlation
Every material test relies on sensors to gather accurate data about the phenomenon under study. The most
common quantities of interest in material
testing are stress and strain. In order to measure these, sensors record the force and deformation
during the test. While force sensing is often as
simple as using a load cell, deformation sensing can be much less straightforward. Traditionally,
deformation was measured using strain gauges and
extensometers. However, these tools come with significant drawbacks and limitations. Recent advances in
optical measurement techniques provide
another way to monitor strain: Digital Image Correlation (DIC). This article will cover the basics of
DIC, including DIC principles, equipment,
algorithms, and output.
Advantages of DIC
Prior to the introduction of DIC, material deformation was usually measured mechanically. This can
be done by attaching a device such as a strain
gauge or an extensometer to a sample. As the sample deforms, the measuring device stretches with it,
producing an electric signal proportional to
the amount of stretching. This signal represents the magnitude of the material deformation over the
area where the sensor is attached.
While mechanical strain sensors can be very accurate, they come with important limitations.
Mechanical strain sensors are best for measuring
strain along a single axis. This is sometimes permissible when the test corresponds to uniaxial
conditions, but strain gauges are inadequate
for biaxial testing such as bulge and FLC because they cannot measure strain along multiple axes in
the same small region.
Another hinderance of strain gauges and extensometers is the fact that they only work locally. A
strain gauge can only measure deformation
happening directly underneath its attached surface. An extensometer can only measure strain that
happens between its two clamping arms.
This means that tests which utilize a large material sample may require mechanical sensors that are
large and heavy, or else multiple
sensors attached over the surface of interest. This can be cost-prohibitive and can interfere with
the test in other ways.
Finally, mechanical sensors are subject to physical limitations. They necessarily take up physical
space near the sample, and this means
they can get in the way of other parts of the testing apparatus. Additionally, some material testing
sample geometries are unsuitable for
sensors that clip on or adhere to their surfaces.
All these restrictions are bypassed by DIC. DIC can measure strain along multiple axes at once, it
can measure strain over the entire sample, and it is not subject to the same physical limitations as
mechanical strain sensors.
Principles of DIC
Digital Image Correlation is an optical measurement technique. It uses cameras to sense and record
movement happening in the real 3D world.
Fundamentally, a camera is a device that collects information from light. The exact pattern of light
reaching a camera sensor is what produces
images, which are essentially packets of stored information. Each image tells us what the world
looked like within the camera field of view at
the time that the image was taken. The key to DIC (as well as other optical measurement techniques)
is to realize that the apparent position
of objects in a camera image can be used to estimate the actual position of those objects in the
This is highly intuitive for large-scale, low-precision measurements. For example, if you see a
photograph where two people appear to be
standing and facing each other, you will probably assume that they were, in fact, standing and
facing each other when the photo was taken.
But why stop with such vague information? Depending on the characteristics of the camera, you might
also be able to estimate the distance
between the tips of their noses to a precision of 0.01 mm.
Similarly, camera images can be used to accurately estimate the position of tiny features on the
surface of a material. Multiple images can be
used to estimate the change in position of those tiny features over time. In other words, cameras
can be used to measure material deformation,
making it possible to use them as strain sensors.
DIC is the method used to accomplish this. It uses a computer algorithm which can be broken down
into three steps: (1) identify key points, (2) track the position of key points in a series of many
sequential images, and (3) interpret the motion of the key points within the images to estimate real
motion in the physical world.
Equipment and Setup
Since DIC is a type of strain sensing, it requires a testing apparatus to gather information
about. Like a traditional extensometer,
DIC systems can be seen as an accessory for a material test; all the other testing equipment is
still required, including testing machines,
force sensors, and computer processing equipment. The exact configuration varies depending on
the type of test.
DIC introduces the additional
limitation that the surface of the material sample must be visible, but for most types of
testing this is easily achieved.
DIC at FADI-AMT
To understand DIC, it is helpful to think in terms of input-process-output. DIC algorithms
require certain information as input. The algorithms
process the input information in a certain way, which creates the desired information (strain,
etc.) as output. The main input required by DIC
is a series of images. These images must be taken sequentially, and they must be taken from the
same location relative to the sample. Therefore,
DIC requires at least one static camera which can record images of the surface of interest in a
defined measurement sequence.
Though DIC can technically be performed with only a single camera, additional pieces of
equipment are often used to make obtaining results easier
and more accurate. Camera accessories are one example of this. Most DIC applications require the
use of camera mounting equipment and additional
lighting. Various types of lenses may be used to achieve the best field of view and distortion
properties; these particulars are beyond the scope
of this article.
DIC at FADI-AMT
Another common DIC practice is to use two cameras instead of one. DIC with one camera is called
2D-DIC, while DIC with two cameras is known as
3D-DIC. With 3D-DIC, paired cameras record images of the same surface. We can then compare image
pairs taken at the same time to obtain more
information than what can be gleaned from a single camera. 2D-DIC and 3D-DIC each have their
advantages. 2D-DIC is less costly, easier to set
up, and easier to use, but requires the camera to be perpendicular to the deforming surface to
give accurate results. 3D-DIC is sometimes more
accurate but is more expensive and often requires lengthy calibration procedures. One major
advantage of 3D-DIC is its ability to measure changes
in depth. While a single camera is unable to detect the distance from the camera to any point of
interest, 3D-DIC provides depth perception by
comparing two images captured from different positions.
One more consideration is crucial: it is important to ensure that the material sample surface
has distinctive features that can be identified and
tracked in multiple images. A flat, uniformly colored surface would provide few identifiable
features. To prevent this, testing engineers apply a
black and white speckle pattern (also called a stochastic pattern) to material samples. This can
be achieved quickly and easily by spray-painting
the material surface to a solid white, then applying a top layer of black paint with a thin
mist. The resulting speckle is high-contrast and
non-uniform, making it easy to detect and identify any point on the surface by tracking a
distinctive region of the paint pattern.
Once a series of images has been taken, the first step of DIC is identify key points. To do
this, DIC software defines a “mesh,” which is a set
of arbitrarily defined points on the material surface in the image. Convention and convenience
dictate that meshes are regular, which means that
the chosen points are evenly spaced.
Once these mesh points are chosen for one image, we need a means to recognize the same points in
other images (in other words, to track them).
To do this, we take a small rectangular region around each point. These regions are called
subsets. Each mesh point has its own subset, with the
mesh point at the center.
Because the sample was painted with a speckle pattern, the appearance of each subset is highly
unique. Thus, we can identify the location of
each subset in each image in the series, and from it we can estimate the location of the
corresponding mesh point.
This process is not as straightforward as it might appear because the material surface (and thus
the paint pattern we use to identify the subset)
is deforming during a test. This causes the subset to appear stretched in later images, which
makes it more difficult to identify.
The heart of DIC algorithms is a mathematical construct called the correlation criterion. The
correlation criterion is a formula that looks at
the difference in appearance between two subsets and quantifies it with a single number.
Comparing two identical subsets gives a correlation
criterion of zero, while comparing very different subsets gives a large correlation criterion.
DIC algorithms track mesh points by taking the original subset and looking for the corresponding
subset in subsequent images. The algorithm
essentially uses a guess-and-check method, taking a subset in the current image (the guess) and
comparing it with the original using the
correlation criterion (the check). It selects the new subset with the lowest correlation
criterion. This method can identify corresponding
subsets even when the material is greatly deformed.
So far, the algorithm has tracked the movement of a set of key points. However, these mesh point
locations still do not represent strain.
The final step in DIC processing is to translate point locations in images (measured in pixels) to
point locations in the real world (with
real distance units). This process is very different for 2D- and 3D-DIC.
2D-DIC uses a simple scale to convert image pixels into physical distances. By taking an image of an
object with known length (such as a ruler),
the known length is measured in pixels. We can then take any pixel distance in the image and
multiply it by the ratio of real length units to
pixels. This process assumes that the surface of the material sample is exactly perpendicular to the
axis of the camera, which is never the
case in the real world. This inevitably creates some measurement error.
3D-DIC does not require a perpendicular camera setup, but the process of converting image distances
to real distances is more complicated.
3D-DIC must be calibrated before use; this involves capturing images of calibration panels or other
well-defined scaling objects in a range
of positions. Various mathematical calibration processes can then be used to determine the relative
orientation of the cameras. With this
information, it is possible to calculate the real-world 3D location of any pair of corresponding
points in a pair of images from the two
cameras. This process is mathematically and computationally intensive compared to the scaling of
2D-DIC, but it is less prone to errors
and provides depth information.
Once the DIC algorithm has calculated the real-world displacement of a set of points on the
material surface, determining the strain is
straightforward. As with any other application, the strain between two points on the material
surface is equal to the change in distance
between them divided by the original distance. Thus, for any pair of mesh points, DIC can
determine the strain between them at any point
in the test.
So far, we have been talking about DIC using discrete points (the mesh). However, DIC is not
limited to measuring deformation/strain only at
these points! In fact, DIC can estimate the deformation at any point on the material surface,
even if it falls between mesh points. This is
done by a process called interpolation, which takes the discrete mesh of displacement values and
estimates the displacement everywhere between
them. This creates a continuous deformation map of the entire surface, allowing us to calculate
strain between any two points on the sample.
Thus, the final product of DIC is a real-time, continuous strain map over the entire surface of a
material testing sample. This provides a
wealth of valuable information about the response of a material to loading. Though DIC is a little
more complicated than using a traditional
strain sensor, the benefits it provides make it a worthwhile addition to many material tests.