We consider a new application of the normalized Hilbert-Huang transform to extract directly the phase from a single fringe pattern. We present a technique to provide, with good accuracy, the phase distribution from a single interferogram without unwrapping step and this by a new exploitation of the analytic signal corresponding to each intrinsic mode function, resulting from one-dimensional empirical mode decomposition of the fringe pattern. A theoretical analysis was carried out for this technique, followed by computer simulations and a real experimental fringe pattern for verification.
1. Introduction
In optical techniques, several physical magnitudes in a variety of scientific and engineering fields are encoded as the phase of a periodic intensity profile. The developments of more sophisticated phase evaluation algorithms are continuously needed [1, 2].
The use of a single fringe pattern for phase retrieval is more suitable in dynamic processes. The unwrapping step, which provides, a continuous phase distribution over its definition domain implies several difficulties and sophisticated algorithms are then needed [3].
Recently, the normalized Hilbert-Huang transform (NHHT) has emerged as a promising and powerful tool for processing nonlinear and nonstationary signals. It consists of three parts: empirical mode decomposition (EMD), intrinsic mode functions (IMF) normalization, and Hilbert spectral analysis (HSA). The NHHT was developed by Huang et al. [4–7] to overcome the limitations set by the Bedrosian and Nuttall theorems when applying the regular Hilbert transform [8–12].
In our work, we attempt to retrieve the unwrapped phase from a single image. Our fringe analysis technique consists of decomposing the fringe pattern into IMFs, applying the HSA to each normalized IMF, then constructing individual analytic signals and computing the phase gradient. This leads directly to the phase distribution avoiding the complex step of phase unwrapping.
Computer simulation results obtained using our method are in very good agreement with those produced using phase-shifting methods [13]. To validate the performance of the proposed technique, it has been further tested on experimental fringe patterns. It was shown that the result totally agrees with that retrieved by phase-shifting techniques.
2. Empirical Mode Decomposition for Fringe Pattern Analysis
The fringe pattern, derived from a two-beam interferometer is characterized by the sinusoidal dependence of the intensity on the spatial coordinates (x,y) of the image plane [1, 2]:
(1)I(x,y)=a(x,y)+b(x,y)cosϕ(x,y),
where a(x,y) is the background. The terms b(x,y) and cosϕ(x,y) are the modulation intensity and the optical phase.
The empirical mode decomposition (EMD) analysis is an adaptive and fully driven data method to decompose any signal into a set of components called intrinsic mode functions (IMF) and a residue [5]. Adding all the IMFs together with the residue reconstructs the original signal without information loss or distortion [14]:
(2)I(x)=∑k=1n-1IMFk(x)+Res(x).
Adding all the IMFs by just skipping the DC component that is, Res(x), we remove the background. The filtered fringe pattern is
(3)J(x)=b(x)cosϕ(x)=∑k=1n-1IMFk(x).
When decomposing the intensity by EMD, the resulting IMFs are not restricted to narrow band signals and can be both amplitude and frequency modulated [8]. Each IMF should be representable in the form
(4)IMFk(x)=bk(x)cos[ϕk(x)],
where the amplitude bk(x) and phase ϕk are both physically and mathematically meaningful. Once a suitable polar parameterization is determined, it is possible to analyse J(x) by processing these individual components. The most common procedure to determine a polar representation is the analytic method of signal processing [15].
3. Analytic Signal Construction
To construct the analytic signal corresponding to each IMF, the simplest way is by employing the Hilbert transform [6]. The real part of an analytic signal is the original input, whereas its imaginary component is the Hilbert transform HT of that signal.
For each IMFk(x), the corresponding analytic signal is
(5)Zk(x)=IMFk(x)+iHT[IMFk(x)].
Using the exponential notation
(6)Zk(x)=bk(x)eiϕk(x)
the following instantaneous attributes of Zk(x) can be defined:
(7)bk(x)={IMFk2(x)+HT[IMFk(x)]2}1/2,(8)IMFNk(x)=IMFk(x)bk(x)(9)ϕk(x)=artctan{bk·HT[IMFNk(x)]IMFk(x)}(10)fk(x)=12πdϕk(x)dx,
where bk(x) is the instantaneous amplitude, IMFNk(x) is the normalized IMFk(x), ϕk(x) is the instantaneous phase, and fk(x) is the instantaneous frequency [16].
In (9), we chose to conduct Hilbert transform on IMFNk(x) rather than on IMFk(x), and this to avoid the Bedrosian and Nuttall restrictions [9–12].
The expansion of the signal in terms of the functions given in (6) is
(11)Z(x)=∑k=1n-1Zk(x)=∑k=1n-1bk(x)eiϕk(x),(12)b(x)eiϕ(x)=∑k=1n-1bk(x)eiϕk(x).
Typically, the modulation term b(x) is a slowly and smoothly varying function compared to ϕ(x), The gradient of (12) leads to:
(13)ib(x)∂ϕ(x)∂xeiϕ(x)=∑k=1n-1[∂bk(x)∂x+i2πfk(x)bk(x)]eiϕk(x).
Dividing (13) by (12) gives then the phase gradient by the real part ℜ of the complex expression in
(14)∂ϕ(x)∂x=R|{∑k=1n-1[2πfk(x)bk(x)-i(∂bk(x)/∂x)]eiϕk(x)}{∑k=1n-1bk(x)eiϕk(x)}|
which leads to the phase by integration of the gradient.
As we can see, the IMFk contribution to the phase gradient is
(15)R|{[2πfk(x)bk(x)-i(∂bk(x)/∂x)]eiϕk(x)}{∑k=1n-1bk(x)eiϕk(x)}|.
4. Numerical Simulations
The numerical simulation consists in generating digitally fringe patterns to verify the ability of the method to determine the phase distribution. The test phase function we used is
(16)ϕ(x,y)=0.0009⋅[(x-128)2+(y-128)2].
The intensity distribution of fringe pattern shown in Figure 1 is
(17)I(x,y)=1+cos(ϕ(x,y)).
We illustrated in Figure 2 the IMF is extracted by EMD from a data row corresponding to y=50.
The fringe pattern I.
The tree IMFs component resulting from EMD decomposition of the data row J(y=50).
As we can see in Figure 3, the physical significance of each IMF is proven since we can see their individual contributions to the global phase phaset.
The individual contribution of each IMF to construct the global phase phaset.
For determining the phase distribution of two dimensional fringe pattern, the previous procedure is generalized to all rows of the image.
The difference map between the simulated phase in Figure 4(a) and the retrieved phase in Figure 4(b) is given in Figure 4(c) where we can see that the error is small in general. Even over the large error location, the error values are smaller than 5%, except for the end region, where the end effect of the Hilbert transform causes additional problems. The occasionally large error values offer an indication where the method failed. All such locations occur at the minimum amplitude. Overall, the NHHT method gives a more consistent, stable quadrature to construct the analytic signal.
(a) Phase gradient in x direction retrieved by (14), (b) phase distribution of (16), and (c) the difference map between the simulated and the retrieved phase.
5. Experimental Results
To illustrate the use of our method for real applications, we tested its performance on a rough ground finish, aluminium surface from a hard disk drive assembly. Changes were induced by tilting the surface and recording interferograms before and after. Figure 5 shows a fringe pattern with 9 fringes of tilt.
Fringe pattern for an aluminium transducer suspension assembly from a hard disk drive that was tilted between measurements.
The 1D empirical mode decomposition method is also applied successively to reduce speckle noise in the correlation fringes produced in digital speckle pattern interferometry (DSPI) [17–19]. When decomposing a data row of the fringe pattern, it is clearly shown in Figure 6 that the first and second IMF contain only speckle noise and must be omitted to construct a smooth fringe pattern.
The four IMFs resulting from EMD decomposition of the data row J(y=110).
The results are of the same order of accuracy to DSPI measurements using the Carré algorithm [20] as it is shown in Figures 7(a) and 7(b). It is worth noting that our method gives the phase distribution with a single image and without unwrapping, while most phase evaluation methods require unwrapping phase over its definition domain, in the manner to provide a continuous phase distribution.
(a) Phase retrieved using our method, and (b) phase retrieved with Carré algorithm.
6. Conclusion
In this paper, we have presented and tested a new method of phase extraction from a single fringe pattern. This was achieved by a new exploitation of the analytic signal corresponding to each normalized intrinsic mode function. Moreover, we gave a physical sense of IMFS derived from fringe pattern decomposition by EMD. We used the NHHT since it has helped to overcome many of the difficulties of the regular Hilbert transform, it gives consistent and stable quadrature when constructing analytic signals. The method applied to simulated and experimental fringe patterns is compared to phase-shifting methods. It shows great robustness and a very close agreement in the computation of the phase.
In summary, the advantages of this technique over other methods include the following.
This method is adequate for real-time acquisition since it provides the optical phase distribution from only one single fringe pattern with a high accuracy.
It is possible to perform an accurate phase demodulation on the fringe pattern even when the image is noisy.
This technique requires less processing time in computation and is easy to implement.
This technique leads directly to the phase distribution avoiding the complex step of phase unwrapping.
The NHHT applied to fringe pattern analysis has defined the qualitative and quantitative contribution of IMFs to recover the phase. This way of fringe pattern analysis is as far as we know has not yet been reported.
Acknowledgments
The authors want to thank Dr. Joseph W. Goodman for his constructive and helpful comments. they also want to thank Dr. Neal Brock and Dr. J. C. Wyant from 4D Technology for providing them the experimental shifted fringe patterns.
SirohiS.ChauS.1999New york, NY, USAMarcel DekkerDorríoB. V.FernándezJ. L.Phase-evaluation methods in whole-field optical measurement techniques1999103R33R552-s2.0-003309724610.1088/0957-0233/10/3/005GhigliaD.PrittM.1998New York, NY, USAJohn Wiley & SonsHuangN. E.ShenZ.LongS. R.A new view of nonlinear water waves: the Hilbert spectrum1999314174572-s2.0-0033489494HuangN. E.ShenZ.LongS. R.WuM. C.SninH. H.ZhengQ.YenN. C.TungC. C.LiuH. H.The empirical mode decomposition and the Hubert spectrum for nonlinear and non-stationary time series analysis199845419719039952-s2.0-5444236478HuangN. E.WuZ.A review on Hilbert-Huang transform: method and its applications to geophysical studies20084622-s2.0-5494914659910.1029/2007RG000228RG2006HuangN. E.LongS. R.2003United States Patent pendingHuangN. E.WuZ.LongS. R.ArnoldK. C.ChenX.BlankK.On instantaneous frequency2009121772292-s2.0-7795166882610.1142/S1793536909000096BedrosianE.A product theorem for Hilbert transform19725128688692-s2.0-0015300021NuttallA. H.On the quadrature approximation to the Hilbert transform of modulated signals19665414581459HuangN. E.BethesdaM.Computing instantaneous frequency by normalizing Hilbert transform2005US Patent 6901353AssidK.DembeleV.alaouiF.NassimA.Normalized Hilbert Huang transform-NHHT applied to phase extraction in wavelet domain20126913CreathK.V phase-measurement interferometry techniques198826C3493932-s2.0-7795697837810.1016/S0079-6638(08)70178-1RillingG.FlandrinP.GoncalvesP.On empirical mode decomposition and its algorithmsProceedings of IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing2003SharpleyR. C.VatchevV.Analysis of the intrinsic mode functions200624117472-s2.0-3364592973910.1007/s00365-005-0603-zBoashashB.Estimating and interpreting the instantaneous frequency of a signal199280520583BerniniM. B.GalizziG. E.FedericoA.KaufmannG. H.Evaluation of the 1D empirical mode decomposition method to smooth digital speckle pattern interferometry fringes20074567237292-s2.0-3394716873810.1016/j.optlaseng.2006.10.007AssidK.DembeleV.alaouiF.NassimA.Bidimensional empirical mode decomposition BEMD applied to speckle denoising for wavelet phase evaluation2011611723FlandrinP.RillingG.GonçalvésP.Empirical mode decomposition as a filter bank20041121121142-s2.0-044232679210.1109/LSP.2003.821662CarréP.Installation et utilisation du comparateur photoélectrique et interférentiel du Bureau International des Poids et Mesures196621, article 00513232-s2.0-000283430410.1088/0026-1394/2/1/005