Вы находитесь на странице: 1из 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267678766

Study and generation of optimal speckle patterns for DIC

Article · January 2007

CITATIONS READS
18 1,431

5 authors, including:

David Lecompte Steven Cooreman


Royal Military Academy ArcelorMittal
50 PUBLICATIONS   964 CITATIONS    49 PUBLICATIONS   656 CITATIONS   

SEE PROFILE SEE PROFILE

Hugo Sol
Vrije Universiteit Brussel
144 PUBLICATIONS   2,780 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Material identification using biaxial testing View project

Single Point Incremental forming View project

All content following this page was uploaded by Hugo Sol on 02 February 2015.

The user has requested enhancement of the downloaded file.


Study and generation of optimal speckle patterns for DIC

D. Lecomptea*, S. Bossuytb , S.Cooremanc, H. Solb, and J. Vantommea


a
Royal Military Academy, Av. de la Renaissance 30, 1000 Brussels, Belgium
b
Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium
c
Technical University KaHo Sint-Lieven, Gebroeders Desmetstraat 1, 9000 Ghent, Belgium

ABSTRACT

Digital Image Correlation (DIC) – also referred to as white light speckle technique – is an optical-numerical full-
field measuring technique, which offers the possibility to determine in-plane displacement fields at the surface of
objects under any kind of loading. For an optimal use of the method, the object of interest has to be covered with
a speckle pattern. The present paper studies the efficiency of a random speckle pattern and its influence on the
measured in-plane displacements with respect to the subset size. Four different speckle patterns are selected
with different speckle sizes and spectral content. Subsequently, each speckle pattern image undergoes a
numerically controlled deformation, which is measured with digital image correlation software. Both imposed and
measured displacements are compared and it is shown that the size of the speckles and its spectral content
combined with the size of the used pixel subset, clearly influences the accuracy of the measured displacements.

INTRODUCTION

The importance attributed to the study of DIC and in this case the speckle pattern is justified by the interest of the
international mechanics of materials community in the use of optical methods for strain measurement. In 2003 a
consortium of eleven organizations from eight different countries began a EU-funded research project
(Standardization Project for Optical Techniques of Strain Measurement - SPOTS) to develop reference tests for
in-plane strain measurement by optical techniques [1]. Their aim is to develop an international standard for optical
strain measurement. This is done to make a stand against the lack of popularity of these powerful techniques
outside the experimental mechanics community. It is tried to provide the users and the developers of such
measurement techniques with a number of tools, which enable them to determine the reliability of the measured
data.A need for calibration and performance evaluation of the optical strain measurement techniques obviously
exists. At the moment, there are no standards to allow calibration to be performed against an internationally
recognized norm.

This paper presents a study on the speckle pattern as a possible error source and its influence on the measured
in-plane displacements with respect to the chosen subset size. It has already been shown [2,3,4] that the subset
size is a critical parameter in the correlation process. To determine the effectiveness of a given speckle pattern, it
will be numerically deformed. The corresponding displacement field is measured by DIC and compared to the
imposed displacements. This approach offers two main advantages: on the one hand it allows dealing with only
one variable, being the aspect of the speckle pattern; on the other hand it allows comparing the displacements
measured by the digital image correlation technique with the numerically imposed displacements, avoiding the
interference of possible variations in the illumination of the pictured surface, instabilities of the experimental set-up
and lens distortion. Recent publications [5,6] reveal the interest of the experimental mechanics community related
to this topic.

Four speckle patterns are considered in this study. First, a speckle pattern is randomly sprayed on a flat white
surface. Several attempts were made to obtain an evenly distributed pattern of black paint spots. An identical
procedure is used in different studies in the case of actual experiments [7,8,9]. Three speckle patterns (figures 1a
,1b and 1c) are obtained by taking different pictures of the same randomly painted speckle pattern, where for
each picture a different zoom is chosen. The obtained speckle patterns are thus different by the size of the
occurring speckles. Additionally, a fourth speckle pattern (figure 1d) is considered in this study. It also shows large
speckles, yet obtained with a very sharp focus, inducing steep grey value edges between the black speckles and
the white background.

Figure 1 - The four considered random speckle patterns obtained by black paint spray

The main difference between the four speckle patterns is thus the overall (and random) size of the black dots and
the spectral content of the corresponding images. It is the influence of these particular aspects that is studied in
the present section. To this end, each image undergoes a numerically controlled deformation, which is measured
by digital image correlation. Both imposed and measured displacements are compared as a function of the subset
size and the type of speckle pattern.

DIGITAL IMAGE CORRELATION

This technique allows studying qualitatively as well as quantitatively the mechanical behaviour of materials under
certain loading conditions and has been used in various technological domains. The DIC technique has been
developed in the 80’s and has since then extensively been evaluated [4,10,11,12] and improved [3,13,14,15].
However, the fundamental principles of the method remain unchanged and are well described in [2,16,17].

Each picture taken with a CCD camera corresponds to a different load step. Two images of the specimen at
different states of deformation are compared by means of a correlation window or an image subset. The image
correlation routine allows locating every subset of the initial image in the deformed image by means of a classic
correlation function using the sum of the squared differences of the pixel values. The displacement result,
expressed in the centre point of the subset, is an average of the displacements of the pixels inside the subset.
The step size defines the number of pixels over which the subset is shifted in x- and y- direction to calculate the
next result. The size of a subset can be for example 9 x 9, 11 x 11, 13 x 13 pixels etc. ; the step size can be 1, 2,
3 pixels etc.
The choice of the correlation window size and the shift or step size is not based on deterministic elements. In the
available commercial DIC-systems the subset size is identical throughout the entire image. This means that once
the subset size is chosen, it remains identical during the correlation calculation. The size of the subset depends
on the type of deformation field to be expected. When dealing with homogeneous deformation (constant
displacement gradients), the subset size should be as big as possible to enable noise filtering and thereby
smoothing of the displacement data. However, in the case of heterogeneous deformations (variable displacement
gradients), the subset size should be a trade-off between smearing out of the displacement data (noise filtering)
and correlation problems. When using large subsets, the displacement data is homogenized in an area that does
not necessarily correspond to the heterogeneous character of the deformation field. When using small subsets,
the grey value pattern is not always unique enough to avoid problems during the correlation calculation.
The step size on the other hand only determines the resolution of available displacement data. The smaller the
step size, the larger the number of displacement data points and the longer the correlation calculation time. The
chosen step size is a compromise between a fair CPU-time and the possibility to treat the displacement data
statistically within a given strain window.
Figure 2 depicts the sequence of taking a picture of an object before and after loading, storing the images onto a
PC trough a frame grabber, performing the correlation of both images – i.e. locating the different undeformed
subsets in the deformed image – and finally calculating the corresponding displacement of the centers of the
subsets, which finally yields the desired displacement field.

Figure 1 – Working principle of the DIC technique

NUMERICAL TRANSFORMATION OF THE IMAGES

The transformation discussed is a vertical displacement between 0 and 1 pixels with shifts of 0.05 pixel. To avoid
interpolation bias, the translations are obtained by imposing the corresponding shift in the Fourier domain as
discussed in [11]. A total of 19 translated images are obtained for every speckle pattern and compared to the
initial image by DIC. The mean and standard deviation of the measured vertical displacement components (in the
± 300 different positions) is calculated for every shifted image. Three subset sizes are used to this end: 9 x 9, 15 x
15 and 31 x 31 pixels. The difference between the measured and imposed displacements is expressed as a
function of the imposed displacements. Figure 3 shows the obtained results for the different speckle patterns.

The entire set of results exhibit a given random error (shown by the standard deviation) as a function of the
displacement value and the speckle pattern used.
It is known that the narrower the correlation peak – indicating the exact position of a subset from the initial image
in the deformed image – the more accurate the position of a subset in the deformed image can be determined. It
is found that the smallest speckles (speckle pattern a) yields the narrowest correlation peak. The larger the
speckles are , the wider the correlation peak is. This ultimately leads to larger random errors on the measured
displacements.
However, figure 3 also shows a larger random error for the smallest speckles than for the medium speckles
(speckle pattern b). This indicates that not only the speckle size influences the random error. It can be observed
that the value of the standard deviation reaches its maximum for a displacement of 0.5 pixel, independently of the
speckle pattern. This is probably the result of the interpolation used during the correlation process, which is
responsible for an amplitude attenuation of the interpolated signal attaining its maximum for a shift of 0.5 pixels,
as discussed [11]. Amplitude attenuation leads to an error in the grey value of the interpolated pixels and
ultimately to a random error in the determination of the position of the shifted subset.

Besides this random error, a systematic sinusoidal error can be observed as well. The systematic error once
again clearly shows the influence of the interpolation mask used in the DIC-software. This error results from a
phase error introduced by the interpolation, needed to perform the correlation calculation [11]. As the positional
error is influenced by the entire frequency content of the image, it is difficult to predict its effect on the measured
displacements. This can only be achieved by studying the systematic error for different speckle patterns. As the
aim of this section is to study the speckle pattern, the influence of the type of interpolation function is not
considered. Although it clearly influences the magnitude of the phase error and amplitude attenuation, the
interpolation mask used is identical for the entirety of the results (as it is a part of the DIC software).

Figure 3 - Errorbar plot of the mean and standard deviation of the difference between imposed and measured
displacement as a function of the imposed displacement, the subset size and the speckle pattern

From the previous considerations it can be concluded that some speckle patterns are more appropriate for image
correlation than others. The aim is now to reveal the differences between the chosen random speckle patterns in
a possibly quantitative manner, by studying their information content in the space domain as well as the spectral
domain.

SPATIAL CHARACTERISATION OF THE SPECKLE PATTERNS

To be able to characterize the speckle patterns more quantitatively, a method to calculate the speckle size
distribution of a given pattern, based on image morphology, will be presented in this paragraph.

Morphology is a technique of image processing based on shapes. The value of each pixel in the output image is
based on a comparison of the corresponding pixel in the input image with its neighbours. By choosing the size
and shape of the neighbourhood, one can construct a morphological operation that is sensitive to specific shapes
in the input image. The operation consists in consecutively eroding and dilating the image. More detailed
information can be found in [18].
In the present case a circular shape is used with different radii. The objective is to locate the speckles with a
radius that is superior to a given value. By performing this operation for different radius values, a speckle size
distribution corresponding to the given pattern can be formulated. This makes it possible to generate a curve,
representing the cumulative speckle covering by speckles within a given radius interval. The result for the different
speckle pattern is shown in figure 4.

Figure 4 - Cumulative percentage of speckles within a given radius interval


for the different random speckle patterns

SPECTRAL CHARACTERISATION OF THE SPECKLE PATTERNS

In a second phase the images of the different speckle patterns are Fourier transformed. Their corresponding
spectral content is shown in figure 5.
Figure 5 clearly indicates the presence of higher wavenumbers for speckle pattern (a) (small speckles) and (b)
(large speckles). Speckle pattern (c) (large speckles as well) exhibits the smallest spectral content. The reason
therefore is that the transition between the dark speckles and the light grey background in speckle pattern (c)
happens smoothly, due to the poor sharpness of the corresponding image.
In summary, small speckles and sharp edges between speckle and background lead to high spectral content,
whereas large speckles and smooth edges between speckle and background limit the spectral content.
Spectral content of the different speckle patterns (a), (b), (c) and (d)

CONCLUSIONS

The combination of the study in the space domain and the study in the Fourier domain yields the following
conclusions:

• On the one hand, small speckles result in a narrow correlation peak, resulting in a more accurate
determination of the subset displacements. Large speckles, on the other hand, lead to a wider correlation
peak, which makes it more difficult to locate the exact position in which the correlation coefficient reaches its
maximum value. This results in a larger random error on the measured displacements.
• Small speckles and speckle patterns exhibiting a steep transition between the dark speckles and the lighter
background, exhibit an important spectral content, i.e. contain higher wavenumbers. This results in a larger
phase error and amplitude attenuation due to interpolation, which translates into a larger random error as well
as a larger systematic error on the measured displacements.
• Speckle pattern (b) containing medium sized speckles and exhibiting a limited spectral content, yields the
most accurate displacement measurements

REFERENCES

[1] E. A. Patterson, E. Hack, P. Brailly, R. L. Burguete, Q. Saleeme,T. Siebert, R. A. Tomlinson, and M. P.


Whelan. Calibration and evaluation of optical systems for full-field strain measurement. Opt.las. Eng.,
doi:10.1016/j. optlaseng.2006.08.012, 2006.
[2] M. A. Sutton, W. J. Wolters, W. H. Peters, W. F. Ranson, and Mc-Neill S. R. Determination of displacements
using an improved digital correlation method. Image and Vision Computing, Vol. 1:133-139,1983.
[3] M. A. Sutton, M. Cheng, W. H. Peters, Y. J. Chao, and S. R. McNeil. Application of an optimized digital image
correlation method to planar deformation analysis. Image Vision Computing, Vol. 4(3):143-150, 1986.
[4] W. G. Knauss, I. Chasiotis, and Y. Huang. Mechanical measurements at the micron and nanometer scales.
Mechanics of Materials, Vol.35:217-231, 2003.
[5] M. Bornert. Resolution and spatial resolution of digital image correlation techniques. In Photomechanics 2006,
book of abstracts, Clermont-Ferrand, France, July 10-12 2006.
[6] N. McCormick, J. Lord, and A. Ive. Towards a fully validated digital image correlation system. In
Photomechanics 2006, book of abstracts, Clermont-Ferrand, France, July 10-12 2006.
[7] J. Abanto-Bueno and J. Lambros. Investigation of crack growth in functionally graded materials using digital
image correlation. Engineering Fracture Mechanics, Vol. 69:1695-1711, 2002.
[8] D. Zhang and D. D. Arola. Applications of digital image correlation to biological tissues. Journal of Biomedical
Optics, Vol. 9(4):691-699, 2004.
[9] S. Samarasinghe and D. Kulasiri. Stress intensity factor of wood from crack-tip displacement-fields obtained
from from digital image processing. Silva Fennica, Vol. 38(3):267-278, 2004.
[10] M. A. Sutton, S. R. McNeill, J. Jang, and J. Babai. Effects of subpixel image restoration on digital correlation
error estimates. Optical Engineering, Vol. 27:870-877, 1988.
[11] H. B. Schreier, J. R. Braasch, and M. A. Sutton. Systematic errors in digital image correlation caused by
intensity interpolation. Optical Engineering, Vol. 39:2915-2921, 2000.
[12] Schreier and M. A. Sutton. Systematic errors in digital image correlation due to under matched subset shape
functions. Experimental Mechanics, Vol. 42:303-310, 2002.
[13] Bruck H.A., McNeil S.R., Sutton M.A. and Peters W.H. Digital Image Correlation using Newton-Raphson
Method of Partial Differential Correction. Experimental Mechanics, 29 (3) 261-267, 1989
[14] P. Cheng, M. A. Sutton, H.W. Schreier, and S. R. McNeill. Full-field speckle pattern image correlation with b-
spline deformation function. Vol. 42:344-352, 2002.
[15] S. Yoneyama and H. Kikuta. Lens distortion correction for digital image correlation by measuring rigid body
displacement. Optical engineering, Vol. 45(2):1-9, 2006.
[16] W. H. Peters and W. F. Ranson. Digital imaging techniques in experimental stress analysis. Optical
Engineering, Vol. 21:427-431, 1982.
[17] T. C. Chu, W. F. Ranson, M. A. Sutton, and W. W. Peters. Applications of digital image correlation techniques
to experimental mechanics. Experimental Mechanics, Vol. 25(3):232-244, 1985.
[18] Matlab. Users Guide, 2000. www.mathworks.com/ products/ optimization/, Version 2.1 (Release12.1), The
Mathworks.

View publication stats

Вам также может понравиться