Digital Image Correlation & Full-Field Measurement
Digital image correlation provides full-field displacement and strain over a measured surface — but the strain is averaged over the subset size, not point strain, and the resolution and uncertainty depend on speckle, calibration, lighting and stereo geometry as much as on the cameras.
What DIC provides
Digital image correlation (DIC) is a full-field optical measurement technique that provides the displacement field — and, by differentiation, the strain field — over a measured surface. Unlike a strain gauge, which gives the strain at one point and in one direction, DIC gives the displacement and strain at every correlated point in the field of view, allowing the engineer to see the deformation pattern, identify the location of the peak strain, and observe features — stress concentrations, strain gradients, crack initiation, buckling modes — that point measurements would miss. The full-field character is the principal advantage of DIC: it turns the measurement from a set of samples into a map. The principal caveat is that the strain field is not point strain; it is averaged over the correlation subset, and its resolution and uncertainty depend on a set of parameters that the engineer must understand and control.
FULL-FIELD DATA DOES NOT REMOVE THE NEED TO UNDERSTAND MEASUREMENT RESOLUTION AND UNCERTAINTY.
How DIC works, conceptually
DIC works by tracking the movement of surface features between images. The specimen surface is prepared with a random speckle pattern — typically a matte white base with a black speckle overlay, applied by spraying, printing or painting. A camera (for 2D DIC) or a pair of cameras (for 3D stereo DIC) images the surface before and during loading. The software divides the reference image into small subsets — square windows of pixels — and searches for the matching subset in each subsequent image, finding the displacement that best correlates the subsets. From the displacement field at every subset centre, the strain field is computed by differentiation over a subgrid of neighbouring displacement points. The stereo arrangement adds a third dimension: by correlating the two camera views, the 3D position of each subset is reconstructed, giving the full 3D displacement field including out-of-plane motion.
DIC parameters and their influence
The quality of a DIC measurement depends on a set of parameters that the engineer controls, and each parameter affects the measurement in a specific way. The table below lists the key parameters, what each affects, how it is optimised, what happens when it is poor, the trade-offs involved, and typical values. DIC is not a black box that produces a strain field from a set of images; it is a measurement system whose performance is determined by these parameters, and a poor choice in any one can degrade the measurement to the point of being misleading.
| Parameter | What it affects | How to optimise | What happens if poor | Trade-offs | Typical values |
|---|---|---|---|---|---|
| Speckle pattern | Correlation quality; ability to track subsets | Random, high-contrast, speckle size 3–5 pixels; uniform coverage | Poor correlation; loss of tracking; noisy displacement | Too fine: aliasing, poor contrast. Too coarse: poor spatial resolution | Speckle size 3–5 pixels; coverage 50% black/white |
| Camera calibration | Accuracy of 3D reconstruction and displacement | Calibrate with a target at the test distance; check residual error | Systematic error in displacement and strain; distortion uncorrected | More images: better calibration but more effort | Residual error < 0.05 pixels for good calibration |
| Stereo angle | Out-of-plane sensitivity; 3D reconstruction quality | Angle of 15–30° between camera axes for general use | Small angle: poor out-of-plane sensitivity. Large angle: poor in-plane | Sensitivity vs correlation robustness | 15–30° for stereo DIC |
| Field of view | Area measured; spatial resolution for a given camera | Choose to cover the region of interest at the required resolution | Too large: coarse spatial resolution. Too small: misses the region | Coverage vs spatial resolution | Determined by camera, lens, and stand-off distance |
| Spatial resolution | Smallest feature the strain field can resolve | Smaller subset and step size give finer resolution | Too coarse: smears strain gradients. Too fine: noisy strain | Resolution vs noise | Subset 15–31 pixels; step 5–10 pixels (typical) |
| Subset size | Correlation robustness; strain averaging volume | Large enough to contain sufficient speckle; small enough to resolve gradients | Too small: noisy, poor correlation. Too large: smears gradients | Robustness vs spatial resolution | 15–31 pixels (typical); depends on speckle |
| Step size | Overlap between subsets; density of displacement points | Smaller step gives denser field; larger step is faster | Too large: sparse field. Too small: redundant computation | Density vs computation time | 5–10 pixels (typical) |
| Lighting | Image contrast; temporal stability | Diffuse, stable, no flicker; avoid specular reflections | Poor contrast: poor correlation. Flicker: apparent strain | Brightness vs heat (can affect specimen) | Diffuse, stable; avoid hot spots and flicker |
The DIC system visualised
The diagram below shows a stereo DIC system. Two cameras view the speckled specimen from different angles. Each camera captures images at defined load steps. The software correlates the speckle pattern between the reference and deformed images, and between the two camera views, to reconstruct the 3D displacement field over the surface. From the displacement field, the strain field is computed by differentiation over a subgrid of neighbouring points. The output is a full-field map of displacement and strain that can be compared to an FE prediction — with the caveat that the DIC strain is averaged over the subset/subgrid, not the point strain of an FE element.
STERO DIC SYSTEM
Camera 1 Camera 2
(left) (right)
│ │
│ stereo angle (15–30°) │
│ ◄──────────────────────► │
│ │
└────────────┬─────────────┘
│
▼
┌───────────────┐
│ speckled │
│ specimen │
│ (random │
│ speckle │
│ pattern) │
└───────────────┘
At each load step:
1. Cameras capture images
2. Software correlates subsets between
reference and deformed images
3. Stereo correlation reconstructs 3D
position of each subset
4. Displacement field → strain field
(by differentiation over a subgrid)
OUTPUT:
• 3D displacement field (u, v, w)
• Surface strain field (εx, εy, γxy)
• Full-field map, comparable to FEStrain from displacement — the averaging issue
The strain field in DIC is computed by differentiating the displacement field over a subgrid of neighbouring displacement points. This means the DIC strain is not the strain at a point; it is the strain averaged over the subgrid used for the differentiation, which in turn depends on the subset size and the step size. A strain gradient that is steep relative to the subgrid is smeared: the peak strain is under-reported and the gradient is broadened. This is not an error in the measurement; it is a property of the method, analogous to the gauge-length averaging of a strain gauge but over a larger area. The consequence is that comparing a DIC strain directly to an FE element strain at a stress concentration can show an apparent discrepancy that is actually a spatial resolution difference: the FE element is small and close to the point value, while the DIC subgrid is larger and averages over a region that includes lower-strain material.
TREATING DIC STRAIN FIELDS AS EQUIVALENT TO FE ELEMENT STRAINS WITHOUT UNDERSTANDING THAT DIC STRAIN IS AVERAGED OVER THE SUBSET/SUBGRID SIZE CAN LEAD TO APPARENT DISCREPANCIES THAT ARE ACTUALLY SPATIAL RESOLUTION DIFFERENCES. DIC strain is not point strain.
Spatial resolution and what it can resolve
The spatial resolution of DIC — the smallest feature the strain field can resolve — is determined by the subset size, the step size and the field of view. A larger field of view at a given camera resolution makes each pixel represent a larger physical area, which coarsens the resolution. A larger subset size increases the averaging volume. A larger step size sparsifies the field. The practical consequence is that DIC cannot resolve a strain gradient that is narrower than a few subsets: the gradient is smeared over the averaging volume. For a smooth strain field, this is not a limitation. For a steep gradient — at a notch root, a crack tip, a ply drop, a contact edge — the peak strain may be under-reported by a significant factor, and the engineer must either reduce the field of view to improve the resolution, or interpret the DIC strain as an averaged value and compare it to an FE strain averaged over the same volume.
Stereo and out-of-plane motion
A single-camera (2D) DIC system measures in-plane displacement on a flat surface, but it cannot distinguish in-plane displacement from out-of-plane motion: if the surface moves towards or away from the camera, the apparent speckle size changes, and the software interprets this as an apparent in-plane strain. A stereo (3D) system with two cameras resolves the out-of-plane motion explicitly, by triangulating the 3D position of each subset from the two views. For tests where out-of-plane motion is expected — bending, buckling, vibration — stereo DIC is essential; 2D DIC will produce apparent strains that are artefacts of the out-of-plane motion. The stereo angle determines the out-of-plane sensitivity: a larger angle gives better out-of-plane resolution but can reduce the in-plane correlation quality, so the angle is a trade-off.
Limitations: lighting, surface, occlusion, noise
DIC has practical limitations that point measurements do not. Lighting must be diffuse, stable and high-contrast; specular reflections create hot spots that corrupt the correlation, and flicker creates apparent strain. The surface must hold a speckle pattern; very smooth, very rough, wet, or hot surfaces may require special preparation. Occlusion — a feature of the rig, a sensor, or a fixture passing between the camera and the surface — creates gaps in the field that must be interpolated or excluded. Noise in the displacement field is amplified by differentiation, so the strain field is noisier than the displacement field, and smoothing or filtering the strain can mask real features. Each of these limitations is manageable, but only if it is recognised; a DIC measurement that looks clean may have been over-smoothed, and a measurement that looks noisy may be showing real strain variation that a gauge would have missed.
DIC and strain gauges together
DIC and strain gauges are complementary, not competitive. DIC gives the full field and can identify the location of the peak strain and the pattern of the strain distribution — information that point measurements cannot provide. Strain gauges give a high-confidence point measurement at a known location, with a well-understood uncertainty and a gauge length that is explicit. A rational measurement strategy for a test where the critical location is not known is to use DIC to find the peak, and to place strain gauges at the peak for a high-confidence point measurement. For a test where the critical location is known, both can be used: DIC for the field and the pattern, gauges for the point. The two measurements should agree at the gauge location; if they do not, the discrepancy is diagnostic — of the DIC spatial resolution, of the gauge bonding, or of the strain field itself.
What to document
A defensible DIC measurement documents the system configuration — cameras, lenses, resolution, stereo angle, stand-off distance, field of view — the speckle pattern and how it was applied, the calibration and its residual error, the analysis parameters — subset size, step size, strain window — and the estimated displacement and strain uncertainty. The spatial resolution — the physical size of a subset and a strain window — should be stated explicitly, so that the strain field can be interpreted as an averaged quantity and compared to FE strains averaged over the same volume. A DIC report that presents a strain field without stating the subset size, the step size and the strain window is presenting a picture whose resolution and uncertainty are unknown, and that is not defensible evidence.