Estimation of phase derivative from a

single fringe pattern using Riesz

transforms

Yassine Tounsi
Sara Zada
Desire Muhire
Ahmed Siari
Abdelkrim Nassim

Yassine Tounsi, Sara Zada, Desire Muhire, Ahmed Siari, Abdelkrim Nassim, “Estimation of phase derivative
from a single fringe pattern using Riesz transforms,” Opt. Eng. 56(11), 111706 (2017),
doi: 10.1117/1.OE.56.11.111706.

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Optical Engineering 56(11), 111706 (November 2017)

Estimation of phase derivative from a single fringe

pattern using Riesz transforms
Yassine Tounsi, Sara Zada, Desire Muhire, Ahmed Siari, and Abdelkrim Nassim*
Chouaib Doukkali University, Measurement and Control Instrumentation Laboratory IMC, Department of Physics, El Jadida, Morocco

Abstract. We propose a technique to estimate the phase derivative in both x and y directions based on Riesz
transform from a single speckle correlation fringes. The originality of this technique is to exploit Riesz transform
for phase derivatives estimation, spatial modulation, speckle denoising, and measure of features similarity.
Phase modulation process is realized by combining a digital spatial carrier and Riesz quadrature; speckle
denoising is computed using Riesz wavelets transform, and the performance is evaluated by Riesz features
SIMilarity. Before applying our method on real speckle correlation fringes, its performance is tested by numerical
simulation. © 2017 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI: 10.1117/1.OE.56.11.111706]
Keywords: Riesz transform; Riesz wavelets transform; phase derivative; digital speckle pattern interferometry.
Paper 170041SS received Jan. 6, 2017; accepted for publication Mar. 28, 2017; published online Apr. 12, 2017.

1 Introduction 2 Theory
The fringe pattern produced in interferometry provides full-
2.1 Riesz Transform
field measurements of physical measurands, such as dis-
placement, strain, temperature, and refractive index changes. In signal processing theory, Hilbert transform is a useful tool
The phase encoded in fringe pattern records values that are for analyzing the one-dimensional (1-D) signal. The linear
directly proportional to the measurands under study. For this combination between a real value function and its Hilbert
reason, fringe analysis has developed gratefully with the transform defines the well-known analytic signal9 from which
development of digital image processing domain to develop we can estimate the local phase, local energy, and local fre-
phase extraction techniques with a good accuracy. A number quency of a 1-D signal.
of techniques have been proposed in the literature for the In the two-dimensional (2-D) case, different approaches
estimation of phase distribution; among these techniques, have been studied in the past years to extend the definition
we cite phase shifting,1 wavelet transform,2,3 and Fourier of Hilbert transform; two of them have gained the greatest
transform.4 Likewise, an extensive research has been carried interest based on multi-dimensional complex analysis and
out on phase derivative estimation due to its ability to pro- Clifford analysis, respectively.
vide the measurement of displacement derivatives. Over the The multidimensional complex analysis defines the direc-
years, a considerable amount of literature has been published tional version of Hilbert transform such as total or multidirec-
for phase derivative estimation, such as techniques based on tional Hilbert transform, whereas Clifford analysis approach
Fourier transform,5 techniques based on difference equation defines another transform called Riesz transform (see
modeling of the interference field,6 and a discrete energy sep- Ref. 8). Contrarily to others extensions of Hilbert transform,
aration algorithm.7 In this context, we propose in this paper a Riesz transform satisfies a steerable property with respect to
method based on Riesz transform considered as multi- rotation and, thus, it is suitable for image analysis.
dimensional extension of the Hilbert transform.8 This The impulse responses of Nth-order complex Riesz oper-
method provides directly from a single modulated fringe pat- ator are defined in spatial representation as10
tern, the phase derivative without requiring any unwrap-
ping step. rn ðx; yÞ ¼ nin ðx þ iyÞn ∕2πðx2 þ y2 Þnþ2∕2 ;
EQ-TARGET;temp:intralink-;e001;326;252 (1)
The second section of this paper presents a generality
about Riesz transform and Riesz wavelet transform; then, in where ðx; yÞ is the spatial coordinate position, n refers to the
the third section, we present the proposed algorithm for order of Riesz operator, and i is the square root of −1.
phase derivatives estimation accompanied with computer Therefore, the first-order 2-D-Riesz kernels along x and y
simulation to evaluate its performance. It will then go on to directions are expressed as
present the principal finding using speckle fringe correlation,
in this section, we briefly introduce the digital speckle pat- rx ¼ x∕2πðx2 þ y2 Þ3∕2 ;
EQ-TARGET;temp:intralink-;e002;326;176

tern interferometry; since the proposed method is sensible to

the noise, we present a new method based on Riesz wavelet ry ¼ y∕2πðx2 þ y2 Þ3∕2 : (2)
for speckle removal and finish by presenting the obtained
results. For a 2-D intensity distribution fðx; yÞ, the first-order
Riesz transform is expressed as
*Address all correspondence to: Abdelkrim Nassim, E-mail: [email protected]
.ma 0091-3286/2017/$25.00 © 2017 SPIE

Optical Engineering 111706-1 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Tounsi et al.: Estimation of phase derivative from a single fringe pattern. . .

Fig. 2 Geometric illustration of Riesz transforms.

qðx; yÞ ¼ −signðtÞ:ðt21 þ t22 Þ1∕2 ;

EQ-TARGET;temp:intralink-;e006;326;626 (6)
Fig. 1 Frequency responses of the components filter of the first-order
Riesz transform, (a) real parts, (b) and imaginary. where t ¼ ðt1 ; t2 Þ and signðtÞ is the signum function defined
as
f R ¼ ðrx þ iry Þ  f ¼ t1 þ it2 ; (3)
signðtÞ ¼ ðt1 ; t2 Þ∕ðt21 þ t22 Þ1∕2 : (7)
EQ-TARGET;temp:intralink-;e003;63;576

EQ-TARGET;temp:intralink-;e007;326;580

where t1 and t2 are the horizontal and vertical Riesz compo- This function is defined according the sign of t, such as
nents of f, respectively, the symbol * stands for the convo- signðtÞ ¼ ð−1; 0; 1Þ, respectively if ðt < 0; t ¼ 0; t > 0Þ.
lution product. From the two Riesz components, Filsberg and Sommer
The easy and fast way to implement Riesz transform is to are constructed the monogenic signal (see Ref. 12) defined
compute its Fourier representation using the 2-D fast Fourier as the linear combination between image and its Riesz com-
transform; in this domain, the filters rx and ry are character- ponents, mathematically, the monogenic signal is formulated
ized by their frequency responses written as as

ðH x ; H y Þ ¼ −ði:u1 ∕kuk; i:u2 ∕kukÞT ;

EQ-TARGET;temp:intralink-;e004;63;476 (4) f monogenic ¼ f þ i:t1 þ j:t2 ;
EQ-TARGET;temp:intralink-;e008;326;471 (8)

where the vector u is composed by u1 and u2 corresponding where i and j mean the orthogonal hypercomplex imaginary
to the frequencies along x and y axes, i is the square root of units. From this monogenic representation, we define three
−1, and T is the transpose operator. monogenic features that characterize image. The local ampli-
In Fig. 1, we display an illustration of 2-D frequency tude that represents the local intensity is defined as
responses of first-order Riesz transform. These frequency qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
responses pairs are associated by a rotation with respect to a ¼ f 2 þ t21 þ t22 :
EQ-TARGET;temp:intralink-;e009;326;395 (9)
each other by 90 deg.
The third way from which we can compute Riesz trans- The local phase represents the structural information, it is
form is the Radon transform,11 that maps 2-D signal onto an defined as
orientation parameterized family of 1-D signal by integrating
the 2-D signal on the line given by the orientation parameter. φðx; yÞ ¼ a tan 2½−signðtÞ:ðt21 þ t22 Þ1∕2 ; f:
EQ-TARGET;temp:intralink-;e010;326;334 (10)
Z þ∞ Z þ∞
EQ-TARGET;temp:intralink-;e005;63;323

The local orientation represents the direction of phase

Rðt; θÞ½fðx; yÞ ¼ fðx; yÞ
−∞ −∞ information, it is defined as
× δðt − x cos θ − y sin θÞdx dy; (5) θðx; yÞ ¼ a tan 2ðt2 ; t1 Þ:
EQ-TARGET;temp:intralink-;e011;326;280 (11)

where R, δ, t, and θ are the Radon transform operator, the Similarly, there is a second-order Riesz transform that is
Dirac delta function, the distance from the origin to the pro- obtained from the first-order Riesz transform components t1
file line, and the angle of direction of the same line, respec- and t2 by
tively. From the Radon space, we can reconstruct the Riesz
transform following the algorithm presented in Refs. 12 t3 ¼ rx ðt1 Þ ¼ rx  ðrx  fÞ;
EQ-TARGET;temp:intralink-;e012;326;216 (12)
and 13.
First, we compute the Radon transform of image,; then, t4 ¼ rx ðt2 Þ ¼ ry ðt1 Þ ¼ ry  ðrx  fÞ;
EQ-TARGET;temp:intralink-;e013;326;186 (13)
for each projection in Radon space, we compute there Hilbert
transform and we multiply on the results by a cos θ and t5 ¼ ry ðt2 Þ ¼ ry  ðry  fÞ: (14)
sin θ. Finally, we obtain the Riesz components by computing
EQ-TARGET;temp:intralink-;e014;326;160

inverse Radon transform.

Geometrically, we represent Riesz transform of an inten- 2.2 Riesz Wavelets Transform
sity distribution f in Cartesian coordinates as shown in Wavelet transform allows a wide application in signal and
Fig. 2. image processing. The idea of wavelet analysis is decompos-
From there, we can define the quadrature map q of f as its ing a signal into a set of basis functions called wavelet. The
projection in the plane ðt1 ; t2 Þ, and the formula of the quad- wavelet’s construction is an outcome by translation and dila-
rature map is written as tation of mother functions ϕ and ψ, called scaling function

Optical Engineering 111706-2 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Tounsi et al.: Estimation of phase derivative from a single fringe pattern. . .

and mother wavelet, respectively, and also called generating transform.18 Riesz wavelet enjoys a greater degree of flex-
functions. ibility for the polar part of the basis functions. In practice,
The multiresolution analysis concept introduced by our wavelet-based multiresolution approach uses a Riesz
Mallat14 was formulated based on the study of the orthonor- Gaussian band-pass pyramid approach. For each pyramid
mal wavelet bases. It has been used for decomposition of level, the original image is first convoluted with a “difference
images into wavelet subspaces. The decomposition needs a of Gaussian” (DOG) kernel and steered using Riesz basis
scaling function that generates a wavelet function. Scaling filters.
function is associated with the impulse response of low- For a wavelet subspace V s , the decomposition of image
pass filter h½n, whereas mother wavelet is associated fðx; yÞ at scale s can be represented as the projection of f
with the impulse response of high-pass filter g½n. The onto V s with the Riesz basis filters Rn .
downsampled outputs of first low-pass and high-pass filters X
provide the approximation (low frequency) from scaling fs ¼ hðψ Rn Þs;k ; I  GDOG
s iðψ̃ Rn Þs;k ;
EQ-TARGET;temp:intralink-;e022;326;642 (22)
coefficients and details (high frequency) from wavelet coef- k∈Z2
ficients, respectively.
pffiffiffiX where GDOG
s is the DOG kernel at scale s and shift k. The
ψðxÞ ¼ 2 g½nψð2x − nÞ;
EQ-TARGET;temp:intralink-;e015;63;598 (15) Riesz basis is given by
n
ðψ Rn Þs;k ¼ 2−s ðψ̃ Rn Þ0;k ð2−s u − kÞ; (23)
pffiffiffiX
EQ-TARGET;temp:intralink-;e023;326;577

ϕðxÞ ¼ 2 h½nϕð2x − nÞ:

EQ-TARGET;temp:intralink-;e016;63;557 (16) with shift k ¼ ðkx ; ky Þ and ðψ̃ Rn Þs;k is the dual basis
n
of ðψ Rn Þs;k .
The filter h½n and g½n are called quadrature mirror filters
and related by the relation
3 Fringe Pattern Phase Derivatives Extraction by
g½n ¼
EQ-TARGET;temp:intralink-;e017;63;498 ð−1Þn h½1 − n: (17) Riesz Transform Algorithm
The intensity distribution of fringe pattern in interferometry
In 2-D case, a 2-D scaling function ϕðx; yÞ, and the three is expressed as
2-D wavelet ΨH ðx; yÞ, ΨV ðx; yÞ, and ΨD ðx; yÞ are critical
elements for wavelet transforms in 2-D.15 These scaling fðx; yÞ ¼ aðx; yÞ þ bðx; yÞ: cos φðx; yÞ;
EQ-TARGET;temp:intralink-;e024;326;467 (24)
function and directional wavelets are composed of the prod-
uct of a 1-D scaling function ϕ and corresponding wavelet ψ, where aðx; yÞ is the bias intensity, bðx; yÞ is the modulation
which are demonstrated as the following: intensity, and φðx; yÞ is the optical phase related directly to
the physical magnitude where their derivative is computed as
ϕðx; yÞ ¼ ϕðxÞϕðyÞ; (18)
ð∇x φ; ∇x φÞT ¼ ð∂φ∕∂x; ∂φ∕∂yÞT :
EQ-TARGET;temp:intralink-;e018;63;401

EQ-TARGET;temp:intralink-;e025;326;403 (25)
ψ ðx; yÞ ¼ ψðxÞϕðyÞ;
EQ-TARGET;temp:intralink-;e019;63;371
H
(19) Eliminating the background aðx; yÞ by low-pass filter, we
obtain an AMFM model19 expressed as
ψ V ðx; yÞ ¼ ϕðyÞψðxÞ;
EQ-TARGET;temp:intralink-;e020;63;346 (20)
^ yÞ ¼ bðx; yÞ: cos φðx; yÞ:
fðx;
EQ-TARGET;temp:intralink-;e026;326;349 (26)
ψ D ðx; yÞ
EQ-TARGET;temp:intralink-;e021;63;320 ¼ ψðxÞψðyÞ; (21)
We generate the quadrature map q of f using first-order
where ΨH , ΨV , and ΨD measure horizontal, vertical, and Riesz transform as presented in the precedent section where
diagonal variations, respectively, from there we obtain hori- we defined the quadrature as defined in Eq. (6).
zontal, vertical, and diagonal details. Combining the AMFM fringe pattern and its obtained
Owing to the uneven angular responses of multiresolution quadrature with a linear carrier characterized with their fre-
analysis, a filter bank with an adaptive orientation property is quency uc and vc along x and y directions, we obtain an
exploited. At each pyramid level, the chosen wavelet trans- intensity distribution of the form
form has to attenuate high-frequency noise. The appropriate
decomposition approach has to be multiscale and multior- f m ðx; yÞ ¼ f:
EQ-TARGET;temp:intralink-;e027;326;239
^ cosðuc x þ vc yÞ
iented. The first multiscale and multioriented image decom-
þ signðtÞ:ðt21 þ t22 Þ1∕2 : sinðuc x þ vc yÞ: (27)
position was performed with the steerable filter banks.16 A
steerable filter has been known for many years and has
This writing means the modulation process realized
been exploited to a variety of features extraction. In an arbi-
digitally.20 Knowing that the term signðtÞ:ðt21 þ t22 Þ1∕2 is the
trary direction, steerable filter is synthesized as a linear com-
quadrature map, we can rewrite f m that represents the modu-
bination of a set of basis filters. Simoncelli and Freeman
lated fringe pattern on the form
suggested combining steerable filters and multiresolution
decomposition in the mid-1990s with their groundbreaking f m ðx; yÞ ¼ b cos φ: cosðuc x þ vc yÞ − b sin φ: sinðuc x þ vc yÞ:
EQ-TARGET;temp:intralink-;e028;326;143

design of the 2-D steerable pyramid.17

In the same context that Simoncelli et al. proposed to (28)
build a steerable pyramid with shift-invariant, translation-
invariant, and rotation-invariant tight wavelet frames with Using trigonometric identity, we obtain the intensity dis-
Riesz transform, that give a called Riesz wavelet tribution of fringes pattern with linear carrier as

Optical Engineering 111706-3 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Tounsi et al.: Estimation of phase derivative from a single fringe pattern. . .

f m ðx; yÞ ¼ bðx; yÞ: cos½uc x þ vc y þ φðx; yÞ;

EQ-TARGET;temp:intralink-;e029;63;752 (29) and second-order Riesz transform features at key locations
between two images (an image with its reference).
where f m means the modulated fringe pattern. In measure- Considering ðt1 ; t2 ; t3 ; t4 ; t5 Þ, the first- and the second-
ment practice, the modulation intensity bðx; yÞ varies slowly order Riesz transforms coefficients of an image f that we
across the fringe pattern. Calculating the derivative of f m , we have presented in the second part. Similarly, the components
obtain ðt1ref ; t2ref ; t3ref ; t4ref ; t5ref Þ are the first- and second-order
Riesz transform coefficients of the reference image f ref .

∇x f m ¼ −b: sinðuc x þ vc y þ φÞ × ðuc þ ∇x φÞ The coefficient RFSIM is defined as
: (30)
∇y f m ¼ −b: sinðuc x þ vc y þ φÞ × ðvc þ ∇y φÞ
EQ-TARGET;temp:intralink-;e030;63;686

Y
n
RFSIM ¼
EQ-TARGET;temp:intralink-;e034;326;664 Di ; (34)
We notice the apparition of term b: sinðuc x þ vc y þ φÞ i¼1
that can be computed by Riesz transform as
where Di is the feature similarity between the two images
b: sinðuc x þ vc y þ φÞ ¼ signðtm Þ:ðt2m1 þ t2m2 Þ1∕2 :
EQ-TARGET;temp:intralink-;e031;63;620 (31) given by
XX XX
Then, we deduce the phase derivatives along x and y Di ¼ di ðx; yÞ:Mðx; yÞ∕
EQ-TARGET;temp:intralink-;e035;326;594 Mðx; yÞ; (35)
directions by
where di is the distance of comparison at the corresponding

∇x φ ¼ ½−∇x f m ∕signðtm Þ:ðt2m1 þ t2m2 Þ1∕2  − uc location ðx; yÞ between the two features ti and tiref (i ¼
: (32) 1; : : : ; 5) is formulated by
∇y φ ¼ ½−∇x f m ∕signðtm Þ:ðt2m1 þ t2m2 Þ1∕2  − vc
EQ-TARGET;temp:intralink-;e032;63;565

di ¼ ½ð2ti :tiref Þ∕ðt2i þ t2iref Þ: (36)

Recall that f m is the intensity distribution of fringe pattern
EQ-TARGET;temp:intralink-;e036;326;526

with linear carrier uc and vc defined previously in Eq. (27).

The feature mask Mðx; yÞ is defined as the edge locations
tm1 and tm2 are the Riesz components of f m . We evaluate
of the each image by
the performance of the proposed method with a simulated
fringe pattern generated by using a phase distribution of Mðx; yÞ ¼ M1  M2 ;
EQ-TARGET;temp:intralink-;e037;326;472 (37)
the form
where  is the logical “OR” operation. M1 denotes the result
φðx; yÞ ¼ 0.0005:½ðx − 256Þ2 þ ðy − 256Þ2 :
EQ-TARGET;temp:intralink-;e033;63;455 (33) of edge detection performed on the image f and M2 is the
result of edge detection on the reference image f ref . The
The chosen phase distribution is a typical example RFSIM metric quantifies how well image details are pre-
because it illustrates a perfect symmetry and present closed served. Comparing the original horizontal and vertical
fringes. Figures 3(a) and 3(b) present a computer simulated phase derivative with their given using the proposed method,
fringe pattern intensity of size (256 × 256 pixels) and its we found an RFSIM value of 0.9261, also, for the obtained
quadrature obtained by Riesz transform, respectively. phase by integration, the RFSIM metric shows 0.9681, this is
The implementation of the proposed method for phase accurate more than the estimated phase by monogenic signal
derivative estimation took about 10 s using MATLAB® with an RFSIM of 0.9345. Noting that the obtained phase
R2010a on a 1.87 GHz Intel Pentium inside the processor derivatives are given for nonzero uc and vc values.
machine.
Figures 4(a) and 4(b) show the estimated phase derivative 4 Application
in both the two directions, and then we can reconstruct the The experimental evaluation of the proposed method is per-
continuous phase distribution easily by implementing a formed with a speckle fringe correlation obtained in speckle
numerical integration. interferometry; it is a powerful optical measurement tech-
To provide a quantitative measure of the resultant images, nique used for industrial measurements to study deforma-
the Riesz transform-based feature similarity (RFSIM) metric tions, vibrations, defects, and damages assessments.22 In
is presented here;21 it allows judging of the image quality experience, speckle pattern exposure of the object is taken
assessment. This metric is computed by comparing the first- in one position. Then the object is deformed, and another

Fig. 3 (a) Simulated fringe pattern and (b) quadrature fringe pattern obtained by Riesz transform,
(c) modulated fringes pattern.

Optical Engineering 111706-4 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Tounsi et al.: Estimation of phase derivative from a single fringe pattern. . .

Fig. 4 The result obtained with numerical simulation. (a) Horizontal estimated phase derivative, (b) ver-
tical estimated phase derivative, and (c) the obtained phase distribution after integration.

exposure is taken. We exploit in this part the speckle fringe for this reason, speckle fringes correlation undergo to a
correlation of fiber carbon given by 4-D Technology denoising step to reduce this noise.
Society®. Figures 5(a) and 5(b) show the recorded speckle To reduce the speckle noise that characterizes the
patterns after and before deformation; these two speckle pat- obtained fringe correlation, we propose in this paper an
terns are subtracted, and their difference is squared to obtain effective method based on Riesz wavelet thresholding tech-
speckle correlation fringes corresponding to the object’s nique. The algorithm is implemented over MATLAB® using
deformation as shown in Fig. 5(c). generalized Riesz-wavelet toolbox for MATLAB®.18 In
Fringes correlation is characterized by a strong speckle Fig. 7, we show the process of the proposed speckle denois-
defined as a granular structure resulting from self interfer- ing algorithm: after Riesz-wavelet decomposition of image
ence of coherent waves randomly scattered from a rough as shown in Fig. 6, the given coefficients undergo threshold-
surface, making it capable of giving the measurement of dis- ing technique, where appropriate threshold limit at each level
placements with an accuracy of order of wavelength used. and threshold method (hard or soft thresholding) are used to
The proposed technique is very sensitive to speckle noise, remove the noise.

Fig. 5 The recorded speckle pattern. (a) After deformation, (b) before deformation, and (c) speckle fringe
correlation.

Fig. 6 Example of Riesz wavelet coefficient of the input image at scales 1, 2, and 3.

Optical Engineering 111706-5 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms

 Tounsi et al.: Estimation of phase derivative from a single fringe pattern. . .

Fig. 7 Process of denoising using Riesz wavelets transforms.

Therefore, the denoised image is obtained by reconstruct- 12. M. Felsberg and G. Sommer, “The monogenic signal,” IEEE Trans.
Signal Process. 49(12), 3136–3144 (2001).
ing thresholded Riesz wavelets coefficients. 13. L. Desbat and V. Perrier, “On locality of radon to Riesz transform,”
After the denoising step, we proceed to applying the pro- Signal Process. 120, 13–25 (2016).
posed technique to estimate phase derivatives using Riesz 14. S. G. Mallat, “A theory for multiresolution signal decomposition: the
wavelet representation,” IEEE Trans. Pattern Anal. Mach. Intell.
transform; the horizontal and vertical phase derivatives are 11(7), 674–693 (1989).
shown in Figs. 7(a) and 7(b), respectively. A 2-D numerical 15. R. C. Gonzalez and R. E. Woods, Digital Image Processing, 3rd ed.,
Pearson, Upper Saddle River, New Jersey (2007).
integration of the two-phase derivative in the two directions 16. W. T. Freeman and E. H. Adelson, “The design and use of steerable
provides the continuous optical phase distribution shown in filters,” IEEE Trans. Pattern Anal. Mach. Intell. 13(9), 891–906 (1991).
Fig. 7(c). 17. E. P. Simoncelli and W. T. Freeman, “The steerable pyramid: a flexible
architecture for multi-scale derivative computation,” in Proc. Int. Conf.
on Image Processing, Vol. 3, pp. 444–447 (1995).
5 Conclusion 18. M. Unser, N. Chenouard, and D. V. D. Ville, “Steerable pyramids and
tight wavelet frames in L2(Rd),” IEEE Trans. Image Process. 20(10),
An estimation procedure based on Riesz transform is intro- 2705–2721 (2011).
duced in this paper for the determination of phase derivative 19. J. P. Havlicek, P. C. Tay, and A. C. Bovik, “AM-FM image models:
and eventually by their integration of phase. It is moreover fundamental techniques and emerging trends,” in Handbook of
Image and Video Processing, A. C. Bovik, Ed., 2nd ed., pp. 377–
interesting to note that the proposed method provides the 395, Elsevier Academic Press, Burlington, Massachusetts (2005).
phase derivative with good accuracy, its performance was 20. E. M. Barj et al., “A digital spatial carrier for wavelet phase extraction,”
Optik–Int. J. Light Electron Opt. 116(11), 507–510 (2005).
substantiated by numerical simulation and judging using 21. L. Zhang, L. Zhang, and X. Mou, “RFSIM: a feature based image qual-
RFSIM metric. In addition, we have applied the proposed ity assessment metric using Riesz transforms,” in 17th IEEE Int. Conf.
method on an experimental speckle fringes correlation. on Image Processing (ICIP ‘10), pp. 321–324 (2010).
22. J. N. Butters and J. A. Leendertz, “Speckle pattern and holographic
Knowing that this technique is sensible to the noise, we techniques in engineering metrology,” Opt. Laser Technol. 3(1), 26–
have proposed a new technique for speckle removal based 30 (1971).
on Riesz wavelets transform.
Yassine Tounsi received his master’s degree in physics of matter
Acknowledgments and radiation in 2015 from Chouaib Doukkali University Morocco.
Currently, he is a PhD student in physics in the same university.
The authors want to thank Dr. Neal Brock and Dr. J. C. His research interest includes wavelets transform, Riesz transform,
Wyant from 4-D Technology for providing them with the monogenic signal and application in digital speckle interferometry,
experimentally shifted fringe patterns. and photoelasticimetry technique.

References Sara Zada received her master’s degree in network and telecommu-
nication from Chouaib Doukkali University Morocco. Currently, she is
1. K. Creath, “V phase-measurement interferometry techniques,” in a PhD student in physics. Her research includes the bidimensional
Progress in Optics, Vol. 26, E. Wolf, Ed., pp. 349–393, Elsevier, empirical mode decomposition, the AMFM model, and applications
Amsterdam, Netherlands (1988). in photoelasticimetry.
2. A. Ghlaifan et al., “Two-dimensional discrete wavelets transform for
optical phase extraction: application on speckle correlation fringes,”
Opt. Eng. 55(12), 121708 (2016). Desire Muhire received his master’s degree in physics of matter
3. M. Afifi et al., “Paul wavelet-based algorithm for optical phase distri- and radiation in 2015. Currently, he is a PhD student in physics at
bution evaluation,” Opt. Commun. 211(1–6), 47–51 (2002). Chouaib Doukkali University, Morocco. His research interest includes
4. M. Takeda, H. Ina, and S. Kobayashi, “Fourier-transform method of AM-FM demodulation for image texture. He is also interested in SAR
fringe-pattern analysis for computer-based topography and interferom- interferometry.
etry,” J. Opt. Soc. Am. 72(1), 156–160 (1982).
5. G. K. Bhat, “A Fourier transform technique to obtain phase derivatives Ahmed Siari received his PhD in physics from Chouaib Doukkali
in interferometry,” Opt. Commun. 110(3), 279–286 (1994). University in collaboration with FIAM Laboratory, Catholic University
6. R. Kulkarni and P. Rastogi, “Simultaneous estimation of phase and
phase derivative using a difference equation representation of the inter- of Louvain, Belgium. During his research career, he published a sev-
ference field,” J. Opt. Soc. Am. A 31(9), 1919–1922 (2014). eral paper in applied physics.
7. R. Kulkarni and P. Rastogi, “Estimation of phase derivatives using dis-
crete energy separation algorithm in digital holographic interferometry,” Abdelkrim Nassim received his PhD in physics from Chouaib
Opt. Lett. 39(13), 3722–3724 (2014). Doukkali University, Morocco, in collaboration with FIAM Laboratory,
8. E. Stein and G. Weiss, Introduction to Fourier Analysis on Euclidean Catholic University of Louvain, Belgium. During his research career,
Spaces, Princeton University Press, Princeton, New Jersey (1971). he published a several paper about speckle interferometry technique
9. D. Gabor, “Theory of communication,” J. Inst. Electr. Eng.–Part I: Gen. in the two principal domains: speckle denoising and optical phase
94(73), 58 (1947).
10. M. Unser and N. Chenouard, “A unifying parametric framework for 2D extraction. He is also a reviewer for JOLT Journal (Optics and Laser
steerable wavelet transforms,” SIAM J. Imag. Sci. 6(1), 102–135 (2013). Technology). He has skills and expertise in speckle interferometry,
11. J. Radon, “On the determination of functions from their integral values wavelet transform, and BEMD decomposition.
along certain manifolds,” IEEE Trans. Med. Imaging 5(4), 170–176
(1986).

Optical Engineering 111706-6 November 2017 • Vol. 56(11)

Downloaded From: http://opticalengineering.spiedigitallibrary.org/pdfaccess.ashx?url=/data/journals/optice/936107/ on 04/13/2017 Terms of Use: http://spiedigitallibrary.org/ss/terms