September 1, 2010 / Vol 35, No 17 / OPTICS LETTERS 2891 High-speed digital-image correlation method: comment Zhaoyang Wang,* Thang M Hoang, Dung A Nguyen, Andrew C Urcinas, and John R Magro Department of Mechanical Engineering, The Catholic University of America, Washington, DC 20064, USA *Corresponding author: wangz@cua.edu Received July 30, 2009; accepted October 5, 2009; posted August 6, 2010 (Doc ID 114561); published August 20, 2010 We comment on the recent letter by Wang et al [Opt Lett 34, 1955 (2009)], in which the authors presented a highspeed digital image correlation (DIC) method We consider that the so-called high-speed DIC method has considerable deficiencies and that the Letter is misleading in terms of applicability and measurement accuracy as well as processing speed © 2010 Optical Society of America OCIS codes: 100.2000, 120.3940, 120.6150, 120.6650 In a recent Letter, Wang et al [1] presented a two-step approach for integer-pixel displacements searching using the digital-image-correlation (DIC) method The method first obtains the correlation index at each pixel with a small-size subset and identifies a group of pixels as potential matching points Then a large-size subset is employed to reanalyze the potential points to find the bestmatching one After that, a peak-finding algorithm (e.g., curved-surface approximation or Lagrange interpolation) is used to get the subpixel displacements This approach is claimed to be a high-speed DIC method We feel that the method has considerable deficiencies and that the Letter seems misleading First, the presented approach in the Letter [1], which comprises two-step integer-pixel displacement searching and subpixel displacement peak-finding, can handle only a deformation field with relatively small rotation or deformation; otherwise, large errors will be expected With respect to applicability and measurement accuracy, the method is inferior to other well-known DIC techniques [2,3] For instance, the iterative cross-correlation algorithm (e.g., the Levenberg–Marquardt method and the Newton–Raphson method [4,5]), which can easily handle large deformation and rotation and provide very high registration accuracy, has been proved to be the most robust DIC algorithm [2,3,6] Second, the Letter [1] seems misleading in regard to the purpose of integer-pixel displacement searching and the processing speed of DIC In practice, DIC normally does not require searching of integer-pixel displacements for all of the pixels defined in the region of interest; instead, the initial estimation needs to be performed only on a starting point After the initial estimation of the starting point, the corresponding subpixel displacements can be subsequently determined by using a popular DIC algorithm, such as the Newton–Raphson method Then, the determined displacements as well as their gradients of the point can be used as the initial estimate of subset parameters for the next point of investigation according to the continuous deformation assumption [3,7] The above handling scheme indicates that even though a “high-speed” integer-pixel displacement-searching approach is useful in some cases, it is helpful only for the analysis of the starting point Accordingly, the total computation time of DIC employing the integer-pixel dis- 0146-9592/10/172891-01$15.00/0 placement-searching scheme will not be evidently reduced In reality, most computation time is consumed by the subpixel registration process, as the image reconstruction at subpixel locations is required Third, the approach proposed in the Letter [1] is very similar to a coarse–fine searching scheme However, a comparison of the two-step searching scheme with the coarse–fine searching algorithms [2,3] was not performed in the Letter [1] On the basis of the testing that we conducted, we did not see a notable advantage of the proposed technique over the existing coarse–fine methods Finally, it may be helpful to point out that an easy and fast, yet very effective, way to perform the initial estimation in DIC is to manually pick one corresponding point (if the rotation and deformation are small) or three corresponding noncollinear points in both the reference and target images [2,3] It has been shown that this human– computer-interaction scheme can provide a reliable initial estimate for very complex deformation fields in practice It is also noteworthy that, using this initial estimation method and the Newton–Raphson algorithm, it usually takes less than s after picking the point pairs to obtain the final subpixel-accuracy displacements for every 1000 points This is considerably faster (also more reliable and practical) than the method presented in the Letter [1], where it takes around s to get the integerpixel displacements and additional time to obtain the subpixel ones at a single point Z Wang acknowledges partial support from the National Science Foundation (NSF) under grant No 0825806 References M Wang, H Wang, and Y Cen, Opt Lett 34, 1955 (2009) B Pan, K Qian, H Xie, and A Asundi, Meas Sci Technol 20, 062001 (2009) M Sutton, in Handbook of Experimental Solid Mechanics, W Sharpe, ed (Springer, 2008), pp 565–600 H Bruck, S McNeil, M Sutton, and W Peters, Exp Mech 29, 261 (1989) G Vendroux and W Knauss, Exp Mech 38, 86 (1998) B Pan, H Xie, B Xu, and F Dai, Meas Sci Technol 17, 1615 (2006) B Pan, Appl Opt 48, 1535 (2009) © 2010 Optical Society of America