Abstract
White light scanning interferometry (WLSI) is a key non-contact three-dimensional (3D) morphometry method. The traditional envelope peak detection (EPD) algorithm is a typical WLSI surface recovery algorithm without phase calculation, but its accuracy is susceptible to noise and fitting errors. In this paper, we propose an improved EPD algorithm based on the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to solve the above problem. The proposed algorithm can accurately locate the position of the envelope peak. The simulation and experiment results demonstrate that the proposed method is effective in improving the accuracy of the measurements.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
The development of precision optics components demands more advanced detection technology [1]. White light scanning interferometry (WLSI) is a mature non-contact optical measurement method with a broad spectrum and short optical coherence length and is a key and widely researched technology [2–4]. The mainstream surface topography solving methods for white light interferometry (WLI) can be divided into two categories: time-domain modulation algorithms based on the envelope amplitude information of the interference signal [5–8] and frequency-domain algorithms based on the signal spectrum phase analysis [9–11].
The time-domain modulation algorithm is mainly used to find the peak of the interferometric signal’s envelope, which is generally obtained by the Fourier transform or Hilbert transform [12,13]. The frequency-domain algorithm gets the phase information of the interference signal by phase shift or Fourier transform algorithm, then the height information is solved by the phase information [10]. The frequency-domain algorithm is generally more accurate than the time-domain modulation algorithm [14], especially the frequency domain algorithm (FDA) proposed by P. de Groot [3]. While the phase calculated by phase shift is susceptible to phase ambiguity, thus additional means are required to correct the phase [15]. The envelope peak detection method (EPD) can avoid this problem because it does not calculate the phase. Paul J. Caber first proposed to determine the envelope peak by fitting simple curves to discrete points near the peak [16]. Min-Cheol Park built on this to determine the envelope peak simply with a single-step matrix multiplication [14]. Larkin proposed to find the peak position of the envelope by fitting a Gaussian function [8]. However, the EPD algorithms select a small number of discrete signals near the peak to fit the envelope curve, which inevitably brings about fitting errors. Furthermore, mechanical vibrations and optical noise during the scanning process, and the overlap of zero-order and second-order fringes caused by low reflections on the surface, can lead to the wrong detection of envelope peaks [17].
In this paper, a new envelope peak detection algorithm is proposed to reduce system noise by the complete ensemble empirical mode decomposition with adaptive noise algorithm (CEEMDAN) algorithm. The CEEMDAN algorithm, proposed by Torres et al. [18], is an improvement of the empirical mode decomposition (EMD) method [19], which decomposes the signal to be processed into a series of single-component signals by adding a moderate amount of Gaussian white noise. Each single-component signal contains only one mode of oscillation, and these decomposed signals are called intrinsic mode functions (IMFs). The noise can be effectively decreased by selecting the IMF closest to the original signal to reconstruct the signal. The proposed algorithm is divided into three steps: firstly, the CEEMDAN algorithm is used to reconstruct the interference signal. Then the FFT algorithm is used to extract the envelope of the reconstructed signal. Finally, the centroid method [10] is used to locate the peak of the envelope. The proposed algorithm can effectively improve the accuracy of envelope peak detection by simulation and experimental verification.
2. Principles
2.1 Interference signals
WLSI is an optical non-contact 3D morphometry method. The interference fringes appear near the position of zero optical path difference (ZOPD) due to the broad spectrum and short coherence length of WLI. During the vertical scan, a series of interferograms are acquired by the CCD. The height information of each pixel point can be obtained by solving the ZOPD position of each pixel point on the interferogram, and then the 3D profile of the sample can be recovered. The intensity of each pixel point on the interferogram can be expressed as [9]:
2.2 Envelope extraction method
The envelope of the WLI signal is generally obtained by the Fast Fourier transform (FFT).
According to Eq. (1), the FFT of the WLI signal can be written as [9]
It can be seen that the envelope curve g(z - h) is directly proportional to the amplitude in Eq. (4). The process of Fourier transform is shown in Fig. 2.
After obtaining the interference signal’s envelope by Fourier transform, the traditional EPD method determines the envelope peak location by selecting a few data points around the maximum value of the envelope function and fitting them by a parabolic or Gaussian function, as shown in Fig. 3. However, the envelope peak position is often incorrectly located by noise and fitting errors, etc. similar to Fig. 1.
2.3 New EPD algorithm with CEEMDAN
CEEMDAN is an improvement algorithm for EMD, which is used to analyze non-linear and non-stationary signals. The EMD has the problem of modal mixing, which can be reduced by adding white noise, but this will introduce extra noise that affects the subsequent analysis and processing of the signal. The CEEMDAN algorithm solves the problem by the following process to mitigate the effect of noise [21]. The algorithm is described as follows [18]:
- (1) Adding N sets of Gaussian white noise signals to the interference signal to get $B(i) = I + \varepsilon \omega (i)$(i = 1, …, N), the first IMF is obtained after EMD decomposition. where I is the interference signal, ω(i) is the Gaussian white noise signal with zero mean and unit variance, ε is the amplitude of the added noise, E is the operator of EMD decomposition, C1 is the first IMF of EMD and r is the residual after EMD decomposition
- (6) The above steps are repeated until the obtained residual signal is a monotonic function, and the EMD decomposition cannot be continued, at which point the algorithm ends.
With the CEEMDAN algorithm, the interference signal is decomposed into a series of IMFs, and we need to select one of them as the reconstructed signal. Calculating the correlation coefficient of each IMF and the original signal, and selecting the IMF with the largest correlation coefficient as the reconstructed signal to reconstruct the 3D shape of the sample. The expression for the correlation coefficient is shown as [22]:
3. Simulation results
In order to verify the effectiveness of the proposed method, a set of simulated white light interference signal data are generated to compare the proposed method with EPD algorithm, the FDA algorithm and the white light phase-shifting interferometry (WLPSI) algorithm. The discrete WLI signal is set to:
We add a random noise with an intensity of 10% of the average light intensity to the simulated ideal interference signal. The FFT transform is used to extract the signal envelope and compare the results of different algorithms. To reduce the uncertainty of the results caused by randomness, each algorithm is simulated 10,000 times and the average of the processing results is calculated. As shown in Table 1, the EPD (P) denotes fitting the envelope by a polynomial and EPD (G) denotes fitting the envelope by a Gaussian function. The simulation results demonstrate that the accuracy of both envelope peak detection methods is improved to some extent by combining the CEEMDAN algorithm. The centroid method combined with CEEMDAN has the smallest error. The FDA has the worst accuracy due to noise, phase unwrapping and then fitting a straight line will get the wrong slope. Due to the additional process of the CEEMDAN algorithm, the time lost increases correspondingly. However, the time consumed is within the acceptable range with MATLAB’s multi-core parallel computing capability.
To further investigate the effect of system noise on the measurement results, six sets of random noise are added to the original interference signal separately. The noise intensities are 5%, 10%, 15%, 20%, 25%, and 30% of the average interference fringe intensity, respectively. As shown in Table 2, the absolute error of the surface height is calculated 10000 times with the different algorithms. The absolute errors of these algorithms become larger as the noise intensity increases. The three envelope peak detection algorithms combined with CEEMDAN feature stronger noise resistance with smaller absolute error at different noise levels The RMS errors of these algorithms at different noise levels are calculated as shown in Table 3. It can be seen that the centroid method with CEEMDAN is the most stable with the smallest RMS error. In general, the above results demonstrate that the centroid algorithm with CEEMDAN can improve the accuracy of envelope peak detection. It should be added additionally that the FDA algorithm has a disproportionately large RMS Error.
4. Experimental results
To further verify the feasibility and effectiveness of the proposed method, the experiments were conducted. The experimental setup included a 10x Mirau interference objective with a numerical aperture of 0.3, a nano-precision piezoelectric transducer (PZT), and a CCD camera with an 8-bit gray level and a resolution of 1024 pixels × 1024 pixels. The light source was an LED with a broad continuous spectrum. The test sample was a standard step block with a height of 7.805 µm. The PZT moved 72 nm per scan, and then the CCD camera recorded the interference fringes and stored them on the computer. The system setup and interferogram are shown in Fig. 7.
The height of the standard steps can be obtained after calculating the height data of the upper and lower surfaces. Ten times tests were conducted with the centroid algorithm combined with CEEMDAN and the results are shown in Table 4. It can be observed that the average step height is 7.829 µm, which is close to the standard step height of 7.805 µm. Furthermore, the standard deviation is tiny, only 0.008 µm.
To compare these algorithms, the sample was tested with the above system. The 3D morphology and 2D cross-sectional profile of the sample are shown in Fig. 8–15, respectively.
It can be seen that using the centroid algorithm to extract the envelope peak can effectively reduce the burr in Fig. 8–13. Due to other factors such as noise, the envelope obtained from the actual tested WLI signal sometimes does not match the single-peak characteristic shown in Fig. 16(a). When the envelope has a bimodal character as shown in Fig. 16(b), the polynomial fit or Gaussian fit will have a large deviation. The centroid method is not overly affected since it is solved by the entire envelope.
Meanwhile, to compare the accuracy of different algorithms, each algorithm was tested with 10 sets of data, the results are shown in Table 5. The experimental results demonstrate that the centroid algorithm with CEEMDAN is more precise and stable compared to the WLPSI and EPD (with/without CEEMDAN) algorithms, which is close to the accuracy of the FDA algorithm. Previous simulations focused on comparing the noise levels that different algorithms can withstand. However, the added noise level is much greater than in actual measurements, so the FDA algorithm does not behave the same way in actual tests as it does in simulations. In fact, the FDA algorithm is extremely accurate in real tests.
5. Conclusion
In this paper, we propose a novel 3D morphology recovery algorithm for WLI based on the CEEMDAN algorithm and centroid method to reduce the error of envelope peak detection. The simulation results demonstrate that the proposed method provides higher accuracy and stronger noise immunity compared to the conventional EPD and WLPSI algorithms. The experimental verification was completed using a sample with a height of 7.805 µm, and the results show that the accuracy of our proposed method is close to that of the FDA algorithm and higher than that of the traditional EPD algorithm and WLPSI algorithm. The simulation and experimental results validate the effectiveness of the method.
Funding
National Natural Science Foundation of China (62175107).
Disclosures
The authors declare no conflicts of interests.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the author upon reasonable request.
References
1. Nikolay I. Zheludev and Yuri S. Kivshar, “From metamaterials to metadevices,” Nat. Mater. 11(11), 917–924 (2012). [CrossRef]
2. P. A. Flournoy, R. W. McClure, and G. Wyntjes, “White-light interferometric thickness gauge,” Appl. Opt. 11(9), 1907–1915 (1972). [CrossRef]
3. P. de Groot and L. Deck, “Surface profiling by analysis of white-light interferograms in the spatial frequency domain,” J. Mod. Opt. 42(2), 389–401 (1995). [CrossRef]
4. P. de Groot, X. C. de Lega, J. Kramer, and M. Turzhitsky, “Determination of fringe order in white-light interference microscopy,” Appl. Opt. 41(22), 4571–4578 (2002). [CrossRef]
5. L. Deck and P. Groot, “High-speed non-contact profiler based on scanning white light interferometry,” Appl. Opt. 35(2), 147–150 (1995). [CrossRef]
6. S. Chen, A. W. Palmer, K. T. V. Grattan, and B. T. Meggitt, “Digital signal processing techniques for electronically scanned optical fiber white-light interferometry,” Appl. Opt. 31(28), 6003–6010 (1992). [CrossRef]
7. T. Guo, L. Ma, J. Chen, X. Fu, and X. Hu, “Microelectromechanical systems surface characterization based on white light phase shifting interferometry,” Opt. Eng. 50(5), 053606 (2011). [CrossRef]
8. K. G. Larkin, “Efficient nonlinear algorithm for envelope detection in white light interferometry,” J. Opt. Soc. Am. A 13(4), 832–843 (1996). [CrossRef]
9. V. Quangsang, F. Fengzhou, Z. Xiaodong, and G. Huimin, “Surface recovery algorithm in white light interferometry based on combined white light phase shifting and fast Fourier transform algorithms,” Appl. Opt. 56(29), 8174–8185 (2017). [CrossRef]
10. L. Zhu, Y. Dong, Z. Li, and X. Zhang, “A Novel Surface Recovery Algorithm for Dual Wavelength White LED in Vertical Scanning Interferometry (VSI),” Sensors 20(18), 5225 (2020). [CrossRef]
11. P. Sandoz, R. Devillers, and A. Plata, “Unambiguous profilometry by fringe-order identification in white-light phase-shifting interferometry,” J. Mod. Opt. 44(3), 519–534 (1997). [CrossRef]
12. C. a. Kino, “Three-dimensional image realization in interference microscopy,” Appl. Opt. 31(14), 2550–2553 (1992). [CrossRef]
13. V. D. Madjarova, H. Kadono, and S. Toyooka, “Dynamic electronic speckle pattern interferometry (DESPI) phase analyses with temporal Hilbert transform,” Opt. Express 11(6), 617–623 (2003). [CrossRef]
14. M. C. Park and S. W. Kim, “Direct quadratic polynomial fitting for fringe peak detection of white light scanning interferograms,” Opt. Eng. 39(4), 952–959 (2000). [CrossRef]
15. G. Young-Sik and D. Angela, “Complete fringe order determination in scanning white-light interferometry using a Fourier-based technique,” Appl. Opt. 51(12), 1922–1924 (2012). [CrossRef]
16. P. Caber, “Interferometric profiler for rough surfaces,” Appl. Opt. 32(19), 3438–3441 (1993). [CrossRef]
17. Z. Lei, X. Liu, L. Chen, W. Lu, and S. J. M. Chang, “A novel surface recovery algorithm in white light interferometry,” Measurement 80, 1–11 (2016). [CrossRef]
18. M. E. Torres, M. A. Colominas, G. Schlotthauer, and P. Flandrin, “A complete ensemble empirical mode decomposition with adaptive noise,” 2011 IEEE international conference on acoustics, speech and signal processing (ICASSP), IEEE (2011), 4144–4147.
19. N. Huang, Z. Shen, and S. Long, “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis,” Proc. R. Soc. Lond. A 454(1971), 903–995 (1998). [CrossRef]
20. I. a. Abdulhalim, “Spatial and temporal coherence effects in interference microscopy and full-field optical coherence tomography,” Ann. Phys. 524(12), 787–804 (2012). [CrossRef]
21. Y Sun, Z Gao, J Ma, et al., “Surface topography measurement of microstructures near the lateral resolution limit via coherence scanning interferometry,” Opt. Lasers Eng. 152, 106949 (2022). [CrossRef]
22. S. Chatterjee, “A New Coefficient of Correlation,” J. Am. Stat. Assoc. 116(536), 2009–2022 (2021). [CrossRef]