Speckle pattern characterisation for high resolution digital image correlation

Speckle pattern characterisation for high
resolution digital image correlation
Mr George Crammond
September 8th 2011
Supervisors: S. W. Boyd and J. M. Dulieu-Barton
Overview & Motivation
Current limitations
Speckle pattern analysis
Numerical image manipulation
Experimental validation
Conclusion
Digital image correlation
LaVision Digital Image Correlation system uses a cross-correlation algorithm to
track the movement of a stochastic speckle pattern on the specimen surface from
a sequence of recorded images
From the algorithm, deformation vectors of the speckle pattern are calculated
and the strains within the material can be evaluated
Reference image t
Deformed image t + dt
Displacement
vector
Deformation vector field
Strain field
Motivation
Image correlation used under high magnification to try and analyse strains
within adhesively bonded joints
Quality of data produced relatively poor
Need to understand the error sources in the calculations to improve data
Complex strain
distributions
Strain / %
High noise
εxx strain in joint loaded at 17kN
Spurious data
DIC Error sources
Lighting
Speckle pattern
Correlation algorithm
Error
Specimen surface
Applied strain
Subset size
Optics
DIC Error sources
Speckle pattern histogram
Speckle size
Lighting
Speckle pattern
Speckle distribution
Number of speckles
Correlation algorithm
Error
Specimen surface
Applied strain
Subset size
Optics
Speckle pattern analysis
Important to consider the speckle pattern due to the physical trade-offs which exist
when conducting digital image correlation
Camera resolution
Pattern uniqueness
No. pixels per speckle
No. speckles per subset
Strain accuracy
Resolution of data
(subset size)
Speckle pattern analysis
Physical properties of patterns and the influence on measurement errors investigated
6.94mm
Testing required to determine the suitability of current speckle patterns under increased
magnification.
Spraycan 296 pixel/mm
Spraycan 296 pixel/mm
Airbrush 296 pixel/mm
Spraycan 705 pixel/mm
Airbrush 705 pixel/mm
2.89mm
8.28mm
Spraycan 705 pixel/mm
3.45mm
Image processing methodology
Morphological approach used to analyse the patterns based upon the shape, size and
distribution of speckles in the pattern
Computer vision techniques required to identify speckles
Laplacian of Gaussian method utilised to provide edge detection of speckles in the
image
Alpha-shape image reconstruction also utilised to repair open contours created from
the edge detection
Image processing methodology
1. Raw image
2. Apply Laplacian
of Gaussian edge
detection method
3. Repair open
contours
using
alpha shape image
reconstruction
4. Produce binary
image from the
detected edges
Pattern evaluation
Pattern type
Spray can
Airbrush
Background colour
Speckle colour
A
X
Black
White
B
X
White
Black
C
X
Black
White
D
X
White
Black
296 pixels/mm
705 pixels/mm
Four different pattern types investigated, altering application method and pattern colour
Patterns tested under two levels of magnification, 296 & 705 pixels / mm
Pattern evaluation
Pattern type
Spray can
Airbrush
Background colour
Speckle colour
A
X
Black
White
B
X
White
Black
C
X
Black
White
D
X
White
Black
705 pixels/mm
296 pixels/mm
A
B
C
D
80
60
40
20
0
0
100
200
300
100
Cumulative percent / %
Cumulative percent / %
100
A
B
C
D
80
60
40
20
0
0
Speckle size / pixels
100
200
Speckle size / pixels
At lower magnification very mixed responses observed
As magnification increases, differences between patterns become greater
Airbrush patterns show more even distribution of speckle sizes
300
Numerical image manipulation
A known displacement field was imposed by manipulating the speckle image in
MatLab in the image Fourier domain
Displacement of stretched images calculated in LaVision DaVis software
Deviation of the imposed and measured strain fields calculated
Numerical image manipulation
705 pixels/mm
296 pixels/mm
0.01
A
B
C
D
0.008
0.006
SD 
SD 
0.008
0.01
0.004
0.002
0
0
A
B
C
D
0.006
0.004
0.002
0.2
0.4
0.6
Strain %
0.8
1
0
0
0.2
0.4
0.6
0.8
1
Strain %
Little difference seen between patterns at lower magnification although distributions different
Under increased magnification, spray can patterns clearly show greater error than the
airbrush
Pattern with a white background also seen to exhibit lower error than those with a black
background
Numerical image manipulation
705 pixels/mm
705 pixels/mm
A
B
C
D
80
60
SD 
Cumulative percent / %
100
40
0.008
A
B
C
D
0.006
0.004
0.002
20
0
0
0.01
100
200
Speckle size / pixels
300
0
0
0.2
0.4
0.6
0.8
1
Strain %
Trends at the higher magnification compliment the distribution analysis conducted earlier
which identified a difference between the pattern properties
Having a pattern dominated by a large number of small speckles appears to be a sub-optimal
solution
Speckle size study
Suspected that the reduced measurement errors for patterns with more even distributions is
linked to the uniqueness of the speckles in the pattern
A random pattern generator was created and different combinations of speckle size and
density investigated
x 10
3
-3
9
2.5
Speckle radius / pixels
8
7
2
6
1.5
5
1
4
3
0.5
2
4
6
8
10
12
14
Speckles per subset
16
18
0
Gradient in error values visible
Identifies lower errors as speckle size and
frequency increase
Larger speckles have a greater level of
uniqueness in size and shape, reducing
uncertainty in measurement
Image processing methodology
Larger speckles also produce a
correlation peak with a wider
footprint
This increases the number of points
which define the peak
Subpixel accuracy improved by this
increase in points due to the 2D
Gaussian curve fit used by LaVision
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
20
40
60
80
100
120
Image processing methodology
Larger speckles also produce a
correlation peak with a wider
footprint
This increases the number of points
which define the peak
Subpixel accuracy improved by this
increase in points due to the 2D
Gaussian curve fit used by LaVision
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
20
40
60
80
100
120
Experimental validation
DIC performed over strain gauge on aluminium specimen at similar magnification levels to
the numerical study
Looking to validate the numerical study results and examine the practical issues with
performing DIC at increased magnification
Experimental validation
296 pixel/mm
-3
02 1.5
x 10
705 pixel/mm
-3
x 10
1.5
StrainAirbrush
gauge0.02
- black
Airbrush
Airbrush
- white
1
0.018
Airbrush
Spraycan
- black
2
- white
SpraySpraycan
can 0.016
18
16
Strain Airbrush
gauge Airbrush
Airbrush 1
Spray Spraycan
can
2
Spraycan
1
Spray can
1
2
0.012
SD 
12
1
Strain / 
0.014
Strain / 
14
01
08
0.01
0.5
0.008
06
0.006
0.5
04
0.004
0
0
500
02
0
0
1000
1500
0
0
2000
500
0.002
Load / N
1000
1500
Variability of results at low magnifications 0for both application methods
1
2
3
4
5
0
Strain %
2000
Load / N
1
2
3
Strain %
4
Difference between application techniques increases under higher magnification
Supports earlier numerical deformation analysis
5
Conclusions
Properties of speckle patterns extensively analysed using a morphological approach
Different pattern types and application methods investigated
Biggest differences seen at the high magnification level
More even distributions of speckle sizes in the pattern appear to have a beneficial
effect on the performance of the pattern
Suspected to be due to an improvement in the subpixel accuracy from larger, more
unique speckles
Overall patterns created on a white background with an airbrush show the best pattern
properties
Any Questions?