Optics and Lasers in Engineering 91 (2017) 178–186

Contents lists available at ScienceDirect

Optics and Lasers in Engineering

journal homepage: www.elsevier.com/locate/optlaseng

Improved FRFT-based method for estimating the physical parameters from MARK
Newton’s rings
⁎
Jin-Min Wua,b, Ming-Feng Lua,b, , Ran Taoa,b, Feng Zhanga,b, Yang Lic
a
School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China
b
Beijing Key Laboratory of Fractional Signals and Systems, Beijing 100081, China
c
Academy of Opto-Electronics, Chinese Academy of Sciences, Beijing 100094, China

A R T I C L E I N F O A B S T R A C T

Keywords: Newton’s rings are often encountered in interferometry, and in analyzing them, we can estimate the physical
Newton’s rings parameters, such as curvature radius and the rings’ center. The fractional Fourier transform (FRFT) is capable of
Fractional Fourier transforms estimating these physical parameters from the rings despite noise and obstacles, but there is still a small
Least squares fitting deviation between the estimated coordinates of the rings' center and the actual values. The least-squares fitting
Measurement
method is popularly used for its accuracy but it is easily affected by the initial values. Nevertheless, with the
Optical metrology
estimated results from the FRFT, it is easy to meet the requirements of initial values. In this paper, the proposed
Fringe analysis
method combines the advantages of the fractional Fourier transform (FRFT) with the least-squares fitting
method in analyzing Newton’s rings fringe patterns. Its performance is assessed by analyzing simulated and
actual Newton’s rings images. The experimental results show that the proposed method is capable of estimating
the parameters in the presence of noise and obstacles. Under the same conditions, the estimation results are
better than those obtained with the original FRFT-based method, especially for the rings’ center. Some
applications are shown to illustrate that the improved FRFT-based method is an important technique for
interferometric measurements.

1. Introduction
Interferometry plays a key role in the field of optical measurements.
As a nondestructive technique, it has been applied in diverse areas of
science and engineering for estimating various physical parameters,
such as curvature radius, wavelength, refractive index, displacement,
and strain. In these applications, the fundamental step is analyzing the
interference fringe patterns, which can be mathematically modeled as
[1]
I (x, y ) = a (x, y ) + b (x, y )cos[φ (x , y )]
where I (x, y ), a (x, y ), b (x, y ), and φ (x, y ) are known as the recorded
intensity, background intensity, fringe amplitude, and phase distribu-
tion, respectively. In general, the information for the measured physical
quantities is encoded in the phase and its derivatives of the recorded
fringe patterns. Therefore, numerous methods for extracting the phase
and its derivatives have been proposed. A widely used technique for
phase measurement is phase-shifting [2]; this technique requires
multiple fringe patterns for analysis, and therefore, it is more suscep-
tible to external disturbances. Thus, a fringe analysis method needed to
be developed that extracted the phase from a single fringe pattern [3,4].
In a similar manner, direct estimation of the phase derivative was also
developed [5–8]. In some of these works, the phase was modeled as
one-dimensional polynomial phase signals (PPS) for each row of the
fringe patterns. By computing the polynomial phase coefficients, the
phase and its derivative could be estimated, and the measured physical
quantities could also be obtained.
In optical metrology, various complex fringe patterns can be
represented using the following two elementary fringe patterns: straight
equispaced fringe patterns and quadratic phase fringe patterns, which
(1) are also called Newton's rings. These elementary fringe patterns are of
great importance because all other fringe patterns can be decomposed
or approximated in terms of these two patterns [9,10]. Therefore,
analyzing these two elementary patterns is very important in inter-
ference fringe analysis. Moreover, Newton's rings are a classical
example of interference fringes, and often encountered when using
interferometers, like Michelson, Twyman-type, and Fizeau-type inter-
ferometers. Newton’s rings are produced when two spherical wave-
fronts interfere due to their different curvature; the rings are character-
ized by circular concentric fringes whose relative spacing becomes
narrower with increasing distance from the center of the pattern. By

⁎
Corresponding author at: School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China.
E-mail address: lumingfeng@bit.edu.cn (M.-F. Lu).

http://dx.doi.org/10.1016/j.optlaseng.2016.11.021
Received 7 August 2016; Received in revised form 23 November 2016; Accepted 23 November 2016
0143-8166/ © 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by/4.0/).
J.-M. Wu et al. Optics and Lasers in Engineering 91 (2017) 178–186

analyzing Newton's rings, one can estimate the curvature radius of

optical components, measure wavelength, detect any displacement and
deformation, and test optical connector end faces.
Various methods for analyzing Newton's rings have been widely
used to measure the physical parameters. A common step in these
methods involves finding the center of the circular rings and computing
the curvature radius. Some methods rely on analyzing the fringe
structure in the interferogram. For instance, the method described in
[11] for extracting fringe skeletons is the classical technique. It is based
on a topological consideration of the interferogram to detect the
positions of the strongest and weakest stripes of the fringes (centerlines
of the fringes), where the value of the fringe intensity is a maximum or
minimum. Then, the measured parameters are obtained through the
fringe skeletons. However, this method is sensitive to noise; therefore, a
variety of filtering methods such as the spin filtering method [12],
wavelet methods [13], and windowed Fourier transform method [3],
have been proposed. In [14], the researchers believed that the
techniques based on topology used relatively less information regarding Fig. 1. Schematic diagram of Newton’s rings.
Newton's rings, and therefore, they presented a new algorithm for
processing these elementary fringe patterns by considering all the pixels
patterns are masked because of the presence of obstacles.
in an image. Because the first step of the algorithm involves finding the
The paper is organized as follows. In Section 2, a mathematical
center of the circular fringes, the curvature-radius estimation errors are
description for Newton’s rings is derived. Then, in Sections 3 and 4, the
generally influenced by the ring center. The aforementioned methods
performances of the FRFT method and the LSF method are presented in
are based on a geometrical approach and require a certain amount of
detail by applying them to some computer-simulated Newton's rings
statistical calculation. In addition to these methods, many studies have
fringe images. Next, in Section 5, the performance of the proposed
also introduced numerical processing methods because the phase
method is assessed with simulated and actual Newton's rings. Its
difference term in the expression for intensity in fringe patterns can
practical applications are also described in this section. Finally, the
be described by a simple quadratic equation [15]，
discussion and conclusions are given in Section 6.
I (x, y ) = I0 + I1⋅ cos(αx x 2 + α y y 2 + ωx x + ωy y + φ0 ), (2)
2. Mathematical description of Newton’s rings
where I0 is the mean intensity in the fringe pattern; I1 is the amplitude;
The basic setup for generating fringe patterns from Newton’s rings is
(αx , α y ) is determined by the rings’ density, which depends on the
shown in Fig. 1. A plano-convex lens is placed on a flat glass, and there
curvatures of the interfering waves; (ωx , ωy ) is related to the coordinates
is an air layer of uneven thickness between the surfaces. When
of the rings' center; and φ0 is the phase. Nascov et al. [16] proposed the
monochromatic light rays are incident on the setup vertically, two sets
least-squares fitting (LSF) method to analyze fringe patterns; their
of reflected light rays are obtained: one is from the curved surface of the
method is implemented with an iterative procedure. The numerical
plano-convex lens, and the other is from the plane surface. When these
algorithm that they proposed determines the fringe parameters with a
two sets of reflected light rays are superposed, they generate inter-
high degree of accuracy; however, the main drawback of the algorithm
ference fringe patterns, commonly known as Newton’s rings; an
is in finding the initial values for the parameters
example is shown in Fig. 2.
(I0, I1 , αx , α y, ωx , ωy , φ0 ). Furthermore, if these initial values are not
Before analyzing the fringe patterns, the intensity distribution of
sufficiently close to the unknown exact values, the iterative procedure
Newton’s rings is considered, which is described by
associated with the method will perform erratically. To overcome this
limitation, Nascov et al. [15] showed that the fringe parameters could I (x, y ) = I0 + I1⋅ cos(∇φ)
be determined using discrete Fourier analysis. Compared with the LSF ⎛ 2π ⎞
= I0 + I1⋅ cos ⎜ λ δ ⎟
method, this method based on the discrete Fourier transform does not ⎝ 0 ⎠
require initial values, but it is as precise as the one based on LSF. ⎛ 2π ⎞
Similarly, in [17], Lu et al. also implemented parameter estimation of = I0 + I1⋅ cos ⎜ λ R r 2 + π ⎟ ,
⎝ 0 ⎠ (3)
optical fringes that did not require initial values for the parameters. The
difference between the two methods, the one proposed in [15,17], lies where I0 and I1 represent the mean intensity and the amplitude of the
in that, in the former, one parameter is determined based on another sine variation of the fringes, respectively; ∇φ is the phase difference and
retrieved parameter, whereas in the fractional Fourier transform (FRFT) ∇φ = φ1 (x, y ) − φ2 (x, y ), where φ1 (x, y ), and φ2 (x, y ) are the phases of
stated in the latter, all fringe parameters describing the quadratic phase the two reflected wavefronts u1 (x, y ) and u2 (x, y ), respectively, which
function are determined simultaneously. Moreover, when some por-
tions of the fringe patterns are blocked out, the FRFT still helps in
effectively estimating the abovementioned parameters. However, the
ring center position obtained by the FRFT approach deviates from the
exact position by small error margins [17].
In this paper, an improved FRFT-based method for analyzing
patterns of Newton's rings is proposed. It combines the advantages of
the FRFT [17] with the LSF algorithm [16] in analyzing Newton’s rings.
This combined method can be used to directly analyze the patterns
without any filter processing, and can provide improved accuracy in
calculating some parameters, especially the position of the ring center.
Furthermore, because of the use of FRFT, our proposed method can be
used to perform fringe analysis even when portions of the fringe Fig. 2. Image of Newton’s rings.

179
J.-M. Wu et al. Optics and Lasers in Engineering 91 (2017) 178–186

are shown in Fig. 1; δ is the path difference of the two reflected light
rays, and according to simple calculation, we can get the relation
δ = r 2 /R + λ 0 /2 , and λ 0 is the incident light wavelength, r is the radii of
Newton’s rings, and R is the radius of curvature of the plano-convex
lens.
Without a loss of generality, it is assumed that the coor-
dinates(x 0 , y0 ) represent the center of the Newton’s rings. By replacing
r 2 by (x − x 0 )2 + (y − y0 )2 and using the definition, 2π /(λ 0 R ) = πK , Eq.
(3) can be modified as follows:
I (x, y ) = I0 + I1⋅ cos[πK (x − x 0 )2 + πK (y − y0 )2 + π ], (4)
cscα⋅ux 0 = ωx , (12)
Fα (ux ) reaches its peak at the location ux0 . Specifically, this indicates
that at the matched rotation angle α , at the location ux0 , the FRFT of the
chirp signal reaches its peak in the FRFT domain. The parameters
(αx , ωx ) can be determined according to Eqs. (10) and (12). The same
calculation is performed on each row of the fringe. Furthermore, the
parameters (α y, ωy ) can also be determined. Once these parameters
are determined, the values of the physical quantities associated with
these parameters can be retrieved. For example, the radius of curvature
of a plano-convex lens R can be obtained using the following formula:

or 4π
R=− ,
λ 0 cot α (13)
I (x, y ) = I0 + I1⋅ cos[πKx 2 − 2πKx 0 x + πKx 02 + πKy 2 − 2πKy0 y + πKy02
and, the center of Newton’s rings can be determined using the following
+ π ] = I0 + I1⋅ cos(αx x 2 + α y y 2 + ωx x + ωy y + φ0 ) (5) equations:
where (αx , α y, ωx , ωy , φ0 ) are constant parameters that describe the ux 0
x0 = ,
intensity distribution of the fringe pattern with the quadratic phase cos α (14)
function. The fringes depicted by this mathematical expression are also
u y0
demonstrated as 2-D chirp signals [18] in the field of signal processing. y0 = .
cos α (15)
From Eq. (5), it can be seen that αx = α y = πK , ωx = −2πKx 0 , and
ωy = −2πKy0 .

3. The fractional Fourier transform 3.2. Experiment

3.1. Methodology Experiments were conducted to analyze the performance of the

FRFTs in estimating the radius of curvature of a plano-convex lens and
As a generalization of the Fourier transform, the FRFT has been used in determining the center of the Newton’s rings. The experiments were
in many applications, such as in-line hologram reconstruction [19], conducted for the cases of noise-free Newton’s rings, blurred Newton’s
image encryption [20,21], optical quantum information processing rings, and Newton’s rings masked by obstacles. The simulated fringes
[22], and interference image enhancement [23]. Moreover, using chirp were created as described in Eq. (4), with the mean intensity I0 = 2 ,
decomposition, it can be used in analyzing fringe patterns with amplitude of the fringes I1 = 2 , wavelength of incident light
quadratic phases (e.g., Newton's rings fringe patterns) [17]. λ 0 = 589.3 nm , curvature radius of the plano-convex lens R = 0.86 m ,
As mentioned in the previous section, the intensity function location of ring center positioned at pixel (256, 256), and the interfer-
I (x, y )is also called a 2-D chirp signal, and each column (or row) of ograms’ size set at 512 × 512 . The simulated noise-free fringe image is
the fringe pattern has the form of a 1-D chirp signal. The 1-D expression shown in Fig. 3(a), and its fractional Fourier spectrum is depicted in
of Eq. (5) is: Fig. 3(b). As shown in Fig. 3(b), we can identify the obvious peak,
I (x ) = rect (x / rm )[I0 + I1⋅ cos(αx x 2 + ωx x + φy )], (6) which is determined by the matched rotation angle α and the optimal
position (ux 0 , u y0 ). Using Eqs. (13)–(15), the value for curvature radius
where [−rm /2, rm /2] is the duration and φy is the fixed phase, which can be obtained, and compared with the exact value, and its relative
varies with the column (or row). errors are about 0.22%; in addition, the retrieved center pixel positions
For simplification, the FRFT of the following complex signal, which are at (254, 254).
is obtained by changing Eq. (6) using a Hilbert transform, is calculated. Furthermore, in the case of a blurred image, we can also identify an
I (x ) = rect (x / rm )⋅I1⋅exp(jαx x 2 + jωx x + jφy ). (7) obvious peak, as shown in Fig. 4. The curvature radius evaluation errors
are found to be 0.41%, and the errors of the estimated center position
The continuous-time FRFT of the 1-D chirp signal is defined as: coordinates, similar to the case of noise-free Newton’s rings, are found
∞ to be 2 pixels. To further verify the capability of noise immunity, some
Fα (ux ) = ∫−∞ Kα (ux , x ) I (x ) dx,
(8) simulated images with different signal-to-noise ratios (SNRs) are used.
These estimated cases can be found in Table 1.
where α is the rotation angle in the time-frequency plane, and Kα (ux , x )
As previously mentioned, not only do we assess the performance of
is the kernel function, which is defined as
the FRFT-based method in the case of white Gaussian noise, but also
⎧ ⎛ 2 2 jux ⎞
check for its robustness in the presence of obstacles. In Fig. 5(a), the
⎪ Bα exp ⎜j u + x cot α − ⎟, α ≠ nπ
⎪ ⎝ 2 sin α ⎠ center of the rings is covered by obstacles and the central regions have
K α (u x , x ) = ⎨ , no fringes. However, the FRFT-based method is still effective because of
⎪ δ (ux − x ), α = 2nπ
⎪ δ (u + x ), the obvious peak in the FRFT domain; this can be seen in Fig. 5(b). In
⎩ x α = 2nπ ± π (9)
such situations, the method obtains the surface curvature with com-
p
where α = 2
0 p 2, n ∈ Z , Bα = (1 − j cot α )/(2π ) .
π, parative errors of approximately 0.22%, which is the same for the ideal
If the rotation angle α is selected using the relation case without obstacles. In addition, the ring center that is obtained is
4π the same as in the ideal case. In Table 2, for various cases of Newton’s
cot α = −2αx = −2πK = − , rings masked by obstacles at different positions, the estimated results
λ0 R (10)
for curvature radius and center location are listed.
then, the expression of Fα (ux ) is a sinc function, i.e., Based on the experimental results, it can be seen that the FRFT-
⎛ cot α 2 ⎞ based method is robust to noise as well as obstacles. Therefore, the
Fα (ux ) = I1 Bα rm exp ⎜j ⋅ux + jφy⎟⋅sinc[(cscα⋅ux − ωx ) rm /2] FRFT-based method, rather than the Fourier-based methods or topol-
⎝ 2 ⎠ (11)
ogy-based methods, is chosen in this paper to calculate the surface
When curvature and the ring center from Newton’s rings fringes.

180
J.-M. Wu et al.
Fig. 3. FRFT of the simulated noise-free Newton’s rings. (a) Simulated noise-free
Newton’s rings. (b) Fractional Fourier magnitude spectrum of the simulated noise-free
Newton’s rings with the matched rotational angle.
4. The least-squares fitting method
4.1. Methodology
The LSF method is popularly used for processing fringes that can be
described mathematically by a function depending on a set of para-
meters. In [16], the principle of this algorithm is discussed in detail.
The basic idea of the algorithm is to find coefficient A to solve the
following problem
2 from the actual location both in the horizontal and vertical directions.
min χ 2 (A) = min I (A, X ) − Y 2 = min ∑ (I (A, Xi ) − Yi )2
A A A
i (16)
where I (A, X ) is the function that represents fringes mathematically, X
is a set of position variables, Y is the input data, and the function χ 2 (A)
represents the sum of the squared normed deviations of the function
I (A, X ) from the input data Y . It is difficult to find an analytical solution
for this problem. Therefore, a numerical solution is needed, which can
be achieved through iteration, for which a set of initial values is
required.
4.2. Experiment
Computer simulations are carried out to demonstrate the validity of
the LSF method. The parameters are selected as follows: the interfer-
ogram size is 512 × 512 pixels; the curvature radius of the plano-convex
lens is 0.86 m ; and the pixel location of the ring center is (256, 256) .
First, we consider the ideal case with a noise-free simulated image.
We assume the initial value of the surface curvature is accurate, and the
181
Optics and Lasers in Engineering 91 (2017) 178–186
Fig. 4. FRFT of the blurred Newton’s rings (SNR=5 dB). (a) Blurred Newton’s rings. (b)
Fractional Fourier magnitude spectrum of the blurred Newton’s rings with the matched
rotational angle.
initial pixel of the rings’ center deviates from the exact value in the
horizontal dimension. The estimated values of the pixel position are
shown in the second column of Table 3. As shown, if the initial location
of the rings’ center is displaced by less than 6 pixels in the horizontal
dimension, the estimated pixel positions are equal to the actual values.
However, when the horizontal deviation is greater than or equal to 6
pixels, the values obtained gradually deviate from the actual values. A
similar conclusion can be drawn for another assumption, which is the
second case where the initial location of the rings’ center is different
Second, we check the performance of the LSF algorithm for the
simulated interferograms, which were blurred by white Gaussian noise
with various SNR values. In Table 4, computational time was recorded
for three cases. It can be found that the computational time decreased
with the decreasing of the difference between the initial values with the
accurate values for the first two cases. However, when the same initial
values are provided for the LSF algorithm, the computational time is
almost same, which is shown in the last column. The simulation results
reveal that the estimated values are not considerably influenced by the
noise level; in contrast, however, they are influenced by the initial
values. Table 5 shows the estimated values obtained by analyzing a
simulated fringe image having SNR = 5. The initial values supplied to
the algorithm in the first case were the same as the initial values used in
the case of the LSF algorithm applied to the ideal case discussed above.
Therefore, Table 5 is the same as Table 3, and consequently, the same
conclusions can be drawn. Furthermore, the estimated values shown in
the last columns of Tables 5 and 3 are computed under the same
condition as the previous case, and the same result is obtained.
J.-M. Wu et al. Optics and Lasers in Engineering 91 (2017) 178–186

Table 1
Estimated results obtained using the FRFT and the improved FRFT vs. SNR.

Estimation Values by use of FRFT Estimation Values by use of improved FRFT

SNR Radius Center Computation Radius Center Computation

(dB) Error (%) Error (pixel) time (s) Error (%) Error (pixel) time (s)

∞ 0.22 (2,2) 990.3 0.22 (0,0) 1066.8

10 0.22 (2,2) 981.7 0.22 (0,0) 1057.1
5 0.22 (2,2) 990.9 0.22 (0,0) 1064.9
0 0.22 (2,2) 983.2 0.22 (0,0) 1055.1
−5 0.41 (2,2) 983.1 0.41 (0,0) 1065.9
−10 1.3 (2,2) 990.0 1.3 (0,0) 1097.4
−15 0.15 (2,8) 990.2 0.15 (1,8) 1056.5
−20 34 (217,221) 992.6 34 (217,221) 1251.0

Table 3
Estimated pixel positions of the rings’ center, obtained with the LSF method for the ideal
patterns.

Deviation in the horizontal dimension Deviation in two directions

Initial Center Estimated Center Initial Center Estimated Center

(Pixel) error (Pixel) (Pixel) error (Pixel)

(256,257) (0,0) (257,257) (0,0)

(256,258) (0,0) (258,258) (0,0)
(256,259) (0,0) (259,259) (0,0)
(256,260) (0,0) (260,260) (0,0)
(256,261) (0,0) (261,261) (3,3)
(256,262) (0,1) (262,262) (5,5)
(256,263) (0,5) (263,263) (7,7)

Therefore, it can be inferred that this algorithm is only sensitive to the

Fig. 5. FRFT of the simulated Newton’s rings with the rings’ center covered by obstacles.
(a) Simulated Newton’s rings covered by obstacles. (b) Fractional Fourier magnitude
spectrum of the simulated Newton’s rings with the matched rotational angle.
initial values provided and not to the level of noise.
To test the dependence of the LSF method on the initial values, we
consider a set of curvature radii with errors and supply these during the
simulation. In addition, the fringes are blurred with a Gaussian noise
with a SNR = 5. As shown in Table 6, if the relative error between the
initial and the actual curvature radius is equal to 1.16%, then the exact
pixels can be obtained when the deviation between the initial pixels and
the exact pixels in the horizontal position is less than or equal to 5
pixels. As the relative error increases, the maximum deviation values of
the horizontal pixel position should decrease in order to obtain an
accurate estimation. Furthermore, when the relative error of initial
curvature radius increases to 3.49%, the exact values are obtained only
when the pixel deviation is less than or equal to 1 pixel deviation.
However, if the difference between the initial and exact pixels in two
directions is 1 pixel, as depicted in the second case of Table 6, then, the
relative error of the initial curvature radius should be less than or equal
to 2.91%. Moreover, if the deviation is equal to 4 pixels in both the
horizontal and vertical directions, we can obtain the exact pixel
positions only when the curvature radius has no error. The same initial
situations have also been analyzed for the noise-free image, and the
same results were obtained.
Based on our experiments, it is clear that the LSF method is not

Table 2
Estimated results obtained using the FRFT and the improved FRFT vs. obstacles.

Estimation Values by use of FRFT Estimation Values by use of improved FRFT

Obstacle Radius Center Computation Radius Center Computation

Position (pixel) Error (%) Error (pixel) time (s) Error (%) Error (pixel) time (s)

(100,100) 0.41 (2,2) 990.4 0.41 (0,0) 1100.3

(200,200) 0.41 (2,2) 982.5 0.41 (0,0) 1090.0
(256,256) 0.22 (2,2) 976.7 0.22 (0,0) 1057.7
(300,300) 0.41 (2,2) 980.0 0.41 (0,0) 1095.4
(400,400) 0.41 (2,2) 983.8 0.41 (0,0) 1096.7

182
J.-M. Wu et al. Optics and Lasers in Engineering 91 (2017) 178–186

Table 4
Computational times corresponding to various initial values for three cases.

Initial Center: (256,256) Initial Radius: 0.83 m Initial Radius: 0.83 m

SNR=∞ SNR=∞ Initial Center: (256,256)

Radius Computational Center Computational SNR Computational

(m) time (s) (Pixel) time (s) (dB) time (s)

0.83 78.8 (256,256) 78.8 ∞ 78.8

0.835 73.8 (256,257) 87.0 15 80.2
0.84 70.4 (256,258) 95.5 10 79.8
0.845 68.3 (256,259) 103.6 5 80.5
0.85 71.0 (257,257) 92.3 0 82.9
0.855 55.9 (258,258) 108.3 −5 79.8
0.86 34.8 (259,259) 119.4 −10 81.8

Table 5 5. The FRFT-LSF method

Estimated pixel position of the rings’ center, obtained with the LSF method for blurred
patterns (SNR=5).
5.1. Methodology
Deviation in the horizontal dimension Deviation in two directions
As previously mentioned, the FRFT method is robust to noise. In
Initial Center Estimated Center Initial Center Estimated Center addition, the FRFT method can be used to estimate physical parameters,
(Pixel) error (Pixel) (Pixel) error (Pixel) such as surface curvature and the center of the rings, even in the
(256,257) (0,0) (257,257) (0,0)
presence of an obstacle in the fringe pattern of the Newton’s rings,
(256,258) (0,0) (258,258) (0,0) especially when the centers are invisible. However, there is still a small
(256,259) (0,0) (259,259) (0,0) deviation between the estimated coordinates of the rings' center and the
(256,260) (0,0) (260,260) (0,0) actual value by about 2 pixels. As mentioned for the LSF method,
(256,261) (0,0) (261,261) (3,3)
however, if the initial values are close to the exact values, the LSF
(256,262) (0,1) (262,262) (5,5)
(256,263) (0,5) (263,263) (7,7) algorithm provides better results. Once the initial values are not
sufficiently close to the exact values, the iterative procedure will
perform erratically or provide results with a high error. In summary,
Table 6 both methods have advantages and disadvantages that are complemen-
Estimated pixel position of the rings’ center, obtained with the LSF method supplied with tary to each other. Therefore, a novel method that combines these two
initial curvature radii with errors for the blurred patterns (SNR=5).
methods is proposed in order to address the disadvantages and obtain
Deviation in the horizontal dimension Deviation in two directions better estimation results.

Radius Initial Center Estimated Initial Estimated Center 5.2. Experiment

Error Center Center
(%) (Pixel) error (Pixel) error (Pixel)
(Pixel) Experiments were conducted to show the efficiency of the improved
FRFT-based method for estimating curvature radius and determining
3.49 (256,257) (0,0) (256,256) (0,0) the rings’ center. The simulated fringe images with various SNR values
2.91 (256,257), (256,258) (0,0) (257,257) (0,0)
are analyzed using the proposed method. The steps in our FRFT-LSF
2.33 (256,257), (256,259) (0,0) (257,257), (0,0)
(258,258) method are as follows:
1.74 (256,257), (256,260) (0,0) (257,257), (0,0)
(258,258) (1) The Newton's rings fringe patterns are processed using the FRFT-
1.16 (256,257), (256,261) (0,0) (257,257), (0,0) based method.
(258,258)
(2) The results in Step 1 are revised by subsequently applying the LSF-
1.16 (256,262) (0,1) (259,259) (1,0)
0.58 (256,262) (0,1) (259,259) (0,0) based algorithm.
0.12 (256,262) (0,1) (260,260) (1,1)
0 (256,262) (0,1) (260,260) (0,0) As shown in the second case of Table 6, when the relative error in
the curvature radius value is 1.74%, the initial center position with a
deviation of 2 pixels or less deviation can be revised by the LSF-based
sensitive to noise, but it is easily influenced by the initial values.
method; these conditions are easy to achieve using the FRFT method. As
Furthermore, the initial values must be given within a certain range
shown in Table 1, when the FRFT is used exclusively, the curvature
depending on the physical property that must be calculated. For
radius error is considerably less than 1.74% in most cases, and the
example, if the rings’ center must be calculated, the initial value
estimated pixels of the rings’ center always maintains a 2-pixel
provided for the curvature radius must be less than or equal to the
deviation, even when the SNR decreases to −10 dB. Thus, the center
critical value shown in Table 6. Accordingly, to calculate the curvature
position can be revised by the subsequent LSF-based method, as shown
radius, the initial value of the rings’ center must be within the deviation
in the last column of Table 1.
values shown in Table 6. These conditions pose a restriction, and hence,
Furthermore, the efficiency of the FRFT-LSF method is shown by
the applicability of the LSF method is limited because of this disadvan-
retrieving the parameters when there is an obstacle in the image. For
tage. Fortunately, however, the initial values can be provided using the
these cases, which are shown in Fig. 6, when the FRFT-based method is
FRFT-based method without these limitations, as discussed in the
used exclusively, the parameters can be retrieved no matter where the
following section.
obstacle is, including the case when the centers are not visible. The
corresponding estimations are recorded in Table 2. It can be seen from
the values in Table 2 that the estimation accuracies of surface curvature
are below 1%, and the deviation of the estimated pixels is 2 pixels.

183
J.-M. Wu et al.
Fig. 6. Simulated Newton’s rings with parts covered by obstacles.
Based on the critical values shown in Table 6, the estimation accuracies
can be improved by the LSF-based method. A better center position
estimate is obtained as shown in the last column of Table 2. Therefore,
it is clear that compared with the FRFT-based method, the improved
method can offer a better estimate of the center’s pixel position
regardless of the fringeless zones and the noise.
To test the proposed method further, three actual interferograms,
which were recorded using a traditional Newton interferometer setup,
are analyzed (shown in Fig. 7). The working wavelength is 589.3 nm,
and the curvature radius of the measured lens is 0.86 m. The camera
has resolution settings of 1600×1200 pixels (pixel size of
6.4 µm×6.4 µm), 1920×1080 pixels (pixel size of 3.9 µm×3.9 µm),
and 1280×720 pixels (pixel size of 5.9 µm×5.9 µm). However, these
pixel images are trimmed to resolutions of 1200×1200, 1080×1080,
and 720×720 for the analysis. Using the FRFT-based method, the
curvature radius is retrieved with an error of approximately 0.34%,
0.9%, and 1.27%. The estimated center positions are, respectively,
(577,453), (487,290), and (325,193). Because the estimated values for
the simulations or actual patterns obtained by the FRFT method meet
the requirements mentioned in the second case of Table 6, the LSF
method is used for correcting the center’s pixel position obtained using
the FRFT method. The revised results are (576,452), (487,289), and
(324,192).
Fig. 7. Schematic of the traditional Newton interferometric setup.
184
Optics and Lasers in Engineering 91 (2017) 178–186
5.3. Applications
5.3.1. Curvature radius estimation of the plano-convex lens
As discussed in Section 5.2, the proposed method can be used with
actual Newton’s rings. The measured curvature radius of the plano-
convex lens was estimated by analyzing the actual interferograms, and
the estimation accuracy was discussed in Section 5.2. Thus, in the
Newton’s rings interferometry, the practical applicability of the pro-
posed method is validated with the experimental results.
5.3.2. Geometric parameter measurements of an optical connector end face
An optical connector is an important component within the fiber
communications field. In order to achieve the best efficiency in optical
communications, an optical connector must be able to perfectly inter-
face two optical fibers together. However, because of the difficulty in
polishing the connector end faces, the polished connector end faces
might not meet the required international standards, and thus, high
connection precision cannot be guaranteed. Therefore, a new method is
required to measure and analyze the geometric parameters of newly
polished connector end faces to ensure high connection precision. In
practice, the parameters marked in Fig. 8 are the key factors that affect
the performance of the connector and the stability of the fiber optic
system. Therefore, it is necessary to measure these key geometrical
parameters accurately to provide polishing details and evaluate their
quality.
To verify the applicability of the proposed method in measuring the
key geometrical parameters, a real fringe pattern (as depicted in Fig. 9)
on the surface of a connector end face is processed. As shown in Fig. 10,
one wavefront is obtained from a reference flat surface (a mirror), and
the other is obtained from the end face to be measured. In general, five
fringe patterns are needed for phase shifting, which is performed by
controlling a piezo transducer. Using the proposed method, a single
fringe pattern can be analyzed to calculate the Radius and Apex-Offset.
The measured optical connector end face is polished in the physical
connect (PC) manner, and a measurement of the angle error is not
required. According to International Electrotechnical Commission (IEC)
J.-M. Wu et al.
Fig. 8. Evaluation parameters of an optical connector end face.
Fig. 9. Interference fringes on the surface of an optical connector end face.
Fig. 10. Michelson interference setup.
specifications, the range of Radius and Apex-Offset is 10–25 mm and 0–
50 µm, respectively. The working wavelength is λ = 650 nm , and the
CCD used has 1024×1024 pixels. For this fringe pattern, the fiber is
located on the central part of the pattern. Using the improved FRFT
method, the Radius of the end face and the center of the fringes can be
determined; then, the Apex-Offset can be determined. Thus, the
proposed method can be applied to measure the Radius and the
Apex-Offset. The measurement results obtained for the Radius and
Apex-Offset are 11.97 mm and 33.35 µm, respectively, for this qualified
optical connector.
6. Discussion and conclusions
In this paper, a new method is proposed to estimate the physical
185
Optics and Lasers in Engineering 91 (2017) 178–186
parameters of a lens in a Newton’s rings fringe-pattern setup, including
curvature radius and the position of the rings’ center. This method
combines the advantages of the FRFT method and the LSF method by
analyzing the Newton’s rings together. Based on our experimental
results, it can be seen that the FRFT method is robust to noise and the
presence of obstacles, but there is a small deviation between the
estimated coordinates of the rings' center and the actual values. In
contrast, we saw that the LSF method can achieve more accurate
estimated results in determining the center positions in the fringe
patterns if the method is provided with a relatively accurate initial
value. Therefore, our improved FRFT-based method combines these
two methods to obtain better estimations than the FRFT-based method
alone. However, compared with optical methods, the relative error for
estimating some parameters, such as curvature radius should be
reduced for high-quality Newton’s rings. Furthermore, fast algorithms
should be introduced to increase the processing speed.
In conclusion, this paper introduces a new approach for extracting
physical parameters from fringe patterns with quadratic phase. The
performance of the proposed method is evaluated by analyzing images
from simulated and actual Newton’s rings. In addition, the applicability
of the proposed method is demonstrated by estimating the curvature
radius of a plano-convex lens and evaluating an optical connector end
face. Furthermore, the proposed method can also be applied to other
similar scenarios. Our future work will involve applying this proposed
method to analyze other quadratic phase fringe patterns. Additionally,
we will address the undercut values in cases measuring the geometrical
parameters of optical connector end faces, and compare the various
methods.
Acknowledgments
This work is supported by the Key Project of the National Natural
Science Foundation of China (Grant no. 61331021), the National
Natural Science Foundation of China for Creative Research Groups
(Grant no. 61421001), and the Foundation for Basic Research of Beijing
Institute of Technology (Grant no. 20140542009). The authors are
grateful to the reviewers for their useful comments.
References
[1] Rajshekhar G, Rastogi P. Fringe analysis: premise and perspectives. Opt Laser Eng
2012;50(8), [III-X].
[2] Creath K. Phase measurement interferometry. In: Wolf E, editor. . Progress in optics,
5. Amsterdam: North-Holland; 1988. p. 349–93.
[3] Qian KM. Two-dimensional windowed Fourier transform for fringe pattern analysis:
principles, applications and implementations. Opt Laser Eng 2007;45(2):304–17.
[4] Rajshekhar G, Gorthi SS, Rastogi P. Estimation of multiple phases from a single
fringe pattern in digital holographic interferometry. Opt Express
2012;20(2):1281–91.
[5] Gorthi SS, Rastogi P. Estimation of phase derivatives using discrete chirp Fourier
transform based method. Opt Lett 2009;34(16):2396–8.
[6] Gorthi SS, Rastogi P. Simultaneous measurement of displacement, strain and
curvature in digital holographic interferometry using high-order instantaneous
moments. Opt Express 2009;17(20):1778491.
[7] Gorthi SS, Rajshekhar G, Rastogi P. Strain estimation in digital holographic
interferometry using piecewise polynomial phase approximation based method. Opt
Express 2010;18(2):560–5.
[8] Kulkarni R, Rastogi P. Phase derivative estimation from a single interferogram using
a Kalman smoothing algorithm. Opt Lett 2015;40(16):3794–7.
[9] Nascov V, Dobroiu A, Apostol D, Damian V. Statistical processing of elementary
fringe patterns. Proc SPIE 2002;5227:43–50.
[10] Nascov V, Dobroiu A, Apostol D, Damian V. Statistical errors on Newton fringe
pattern digital processing. Proc SPIE 2004;5581:788–96.
[11] Zhang F, Wang DY, Xiao ZT, Geng L, Wu J, Xu ZB, et al. Skeleton extraction and
phase interpolation for single ESPI fringe pattern based on the partial differential
equations. Opt Express 2015;23(23):29625–38.
[12] Tang C, Wang LL, Yan HQ, Li CC. Comparison on performance of some represen-
tative and recent filtering methods in electronic speckle pattern interferometry. Opt
Laser Eng 2012;50(8):1036–51.
[13] Federico A, Kaufmann GH. Comparative study of wavelet thresholding methods
techniques for denoising electronic speckle pattern interferometry fringes. Opt Eng
2001;40(11):2601–4.
[14] Dobroiu A, Alexandrescu A, Apostol D, Nascov V, Damian V. Centering and
J.-M. Wu et al. Optics and Lasers in Engineering 91 (2017) 178–186

profiling algorithm for processing Newton's rings fringe patterns. Opt Eng
2000;39(12):3201–6.
[15] Nascov V, Apostol D, Garoi F. Statistical processing of Newton's rings using discrete
Fourier analysis. Opt Eng 2007;46(2):028201.
[16] Nascov V, Dobroiu A, Apostol D, Damian V. Automatic digital processing of
Newton's rings fringe patterns. Proc SPIE 2000;4430:835–41.
[17] Lu MF, Zhang F, Tao R, Ni GQ, Bai TZ, Yang WM. Parameter estimation of optical
fringes with quadratic phase using the fractional Fourier transform. Opt Laser Eng
2015;74:1–16.
[18] Lu MF, Ni GQ, Bai TZ, Tao R, Zhang F. Method for suppressing the quantization
error of Newton's rings fringe pattern. Opt Eng 2013;52(10):103105.
[19] Zhang Y, Pedrini G, Osten W, Tiziani HJ. Applications of fractional transforms to
object reconstruction from in-line holograms. Opt Lett 2004;29(15):1793–5.
[20] Zhang F, Hu Y, Tao R, Wang Y. New fractional matrix with its applications in image
encryption. Opt Laser Technol 2014;64:82–93.
[21] Li YB, Zhang F, Li YC, Tao R. Asymmetric multiple-image encryption based on the
cascaded fractional Fourier transform. Opt Laser Eng 2015;72:15–25.
[22] Sun Y, Tao R, Zhang XD. Fractional Fourier processing of quantum light. Opt
Express 2014;22(1):727–36.
[23] Lu MF, Zhang F, Tao R, Ni GQ, Bai TZ. Application of fractional Fourier transform
for interferometry. In: Proceedings of the conference on lLasers and electro-optics.
OSA Technical Digest (online). Optical Society of America. 2015. paper JW2A.104.

186