Microscopy, Volume Advance Article (3) – Mar 26, 2018

/lp/ou_press/phase-retrieval-using-through-focus-images-in-lorentz-transmission-Rg9DRbvZv0

- Publisher
- Oxford University Press
- Copyright
- © The Author(s) 2018. Published by Oxford University Press on behalf of The Japanese Society of Microscopy. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
- ISSN
- 0022-0744
- eISSN
- 1477-9986
- D.O.I.
- 10.1093/jmicro/dfy014
- Publisher site
- See Article on Publisher Site

Abstract In this study, we report on phase retrieval by the maximum likelihood method with conjugate gradient method (CG-MAL) using through-focus images in Lorentz transmission electron microscopy (LTEM). The method was evaluated using 32 simulated and experimentally obtained through-focus images of magnetic bubbles; these images were collected under a defocus range from approximately −3 mm to 3 mm. Consequently, we obtained the magnetic domain structures of the magnetic bubbles in both the simulation and the LTEM experiment. Furthermore, the CG-MAL method showed better convergence behavior than other iterative phase retrieval methods. Therefore, the method can also be widely and effectively applied to the observation of magnetic domain structures other than magnetic bubbles when highly defocused through-focus images are used. Lorentz transmission electron microscopy, phase retrieval, focal-series reconstruction, iterative wave function reconstruction, maximum likelihood method, magnetic domain structure Introduction In Lorentz transmission electron microscopy (LTEM) [1], phase retrieval using through-focus images [2] is a powerful technique for observing magnetic domain structures within samples in real space. Although this approach has the advantage of enabling phase retrieval without any additional devices, it also has several disadvantages compared with other phase retrieval techniques. For instance, the time resolution of the focal-series reconstruction methods is usually lower than that of techniques such as differential phase contrast imaging in scanning TEM (DPC-STEM) [3, 4] because the focal-series reconstruction methods usually require multiple defocused images and image processing. More importantly, the quantitative performance of the focal-series reconstruction methods is often inferior to that of off-axis electron holography [5, 6]. Performing the focal-series reconstruction more quantitatively requires utilizing highly defocused images in a wide defocus range because the magnitude of the observed image contrast generally depends on the magnitude of the defocus value. A phase retrieval method that utilizes a transport-of-intensity equation (TIE) [7] (i.e. a TIE method) is the most popular focal-series reconstruction method in LTEM; it generally involves an in-focus image and two defocused images [8]. However, the TIE method has several problems attributed to the fact that the method can appropriately deal with only three LTEM images obtained with a sufficiently small focal interval [9]. In the TIE method, the phase contrasts transferred to the three LTEM images are too weak for the phase distributions to be determined accurately. The phase distributions obtained by the TIE method have been reported to often suffer from noise amplification and image artifacts [10], which makes observing detailed magnetic domain structures difficult. Therefore, to perform precise and accurate focal-series reconstruction, it is necessary to apply a method that can appropriately utilize numerous highly defocused through-focus images. In high-resolution TEM (HRTEM), various focal-series reconstruction methods have been developed in addition to the TIE method. These focal-series reconstruction methods can generally be classified into two types: Fourier analytical reconstruction methods and iterative reconstruction methods. The three-dimensional Fourier filtering method (3DFFM) [11, 12], paraboloid method [13] and real-time wave field reconstruction method [14, 15] are classified as Fourier analytical reconstruction methods. By contrast, the iterative reconstruction methods include the iterative wave function reconstruction (IWFR) method [16], the maximum likelihood method with conjugate gradient method (CG-MAL) [17] and the maximum likelihood method with the steepest descent method (SD-MAL) [18]. Compared with the TIE method, these focal-series reconstruction methods can make use of numerous through-focus images collected under a larger defocus range. In particular, the iterative reconstruction methods can appropriately retrieve low-spatial-frequency components in phase distribution, as previously reported [19], enabling reconstruction of the phase information in a quantitative manner in a wide Fourier spatial frequency range. Therefore, we also expect that they can be used for phase retrieval that utilizes numerous highly defocused through-focus in the LTEM. In this study, we report phase retrieval by the CG-MAL method using through-focus images in the LTEM. As an application example in this study, we applied the method to magnetic bubbles [20, 21]. We evaluated the method using simulated through-focus images of a magnetic bubble, where the images were calculated under the same defocus conditions as those used in an experiment. Subsequently, the phase distribution was retrieved using 32 experimentally obtained through-focus images of magnetic bubbles, which were obtained under a defocus range from approximately −3 mm to 3 mm. The results indicate that the CG-MAL method is particularly promising for detailed analyses of magnetic domain structures using several largely defocused through-focus LTEM images. Theory In this section, we describe the theoretical background and procedures of the CG-MAL method. In the MAL method, phase retrieval is performed by an iterative process. We represent the reconstructed electron wave function after j iterations of the MAL method as ϕx,y(j)(j=0,1,2⋯). The wave function reconstructed by the Fourier analytical reconstruction methods is usually used as the initial wave function ( ϕx,y(0)). In this study, we obtained the initial wave function by 3DFFM [11]. An advantage of using the 3DFFM is that various parameters such as the temporal partial coherence parameter and the linear sample drift are directly measured and evaluated from the 3D Fourier spectrum [22, 23]. Using the known defocus values of the original through-focus images, we can computationally calculate the through-focus images from the reconstructed wave function ( ϕx,y(j)) using Eq. (1) : Ix,y,Δf′(j) (ϕx,y(j))=|ϕx,y(j)⊗tx,y,Δf|2, (1) where x, y and Δf denote the 2D coordinates and the defocus value, respectively, tx,y,Δf denotes the point spread function [18] at each focus position and ⊗ denotes the convolution operator. At this point, we define the difference between the intensities of the original images and computationally calculated images as an error function: E(j)(ϕx,y(j))=1NΔf∑Δf(∑x,y|Ix,y,Δf−Ix,y,Δf′(j)(ϕx,y(j))|2∑x,yIx,y,Δf), (2) where Ix,y,Δf and Ix,y,Δf′(j) are the original and the computationally calculated through-focus images, respectively, NΔf denotes the total number of through-focus images and ∑ denotes a summation operator. The reconstructed wave function for the next iteration ( ϕx,y(j+1)) is calculated such that the current error function ( E(j)) is minimized using convex optimization techniques such as the conjugate gradient method [17] or steepest descent method [18]. These methods are termed the CG-MAL and SD-MAL methods, respectively. The basic idea behind the MAL method is that the error function returns zero when the true wave function is obtained. That is, the true wave function minimizes the error function. The iterative process continues until the error function sufficiently converges. The updated wave function is obtained based on the following equation: ϕx,y(j+1)=ϕx,y(j)−α·dx,y(j)(α>0), (3) where dx,y(j) denotes the search direction and α denotes a feedback parameter that minimizes the error function as much as possible along the search direction. In the case of the SD-MAL method, the search direction ( dx,y(j)) equals the gradient direction of the error function. In the case of the CG-MAL method, the search direction is determined from the gradient direction of the error function and the previous search direction ( dx,y(j−1)) using the Polak–Ribiére formula [17]. The gradient direction of the error function is expressed as gx,y(j)=−1NΔf∑Δf[{(Ix,y,Δf−I′x,y,Δf(j)(ϕx,y(j)))⋅(ϕx,y(j)⊗tx,y,Δf⁎)}⊗tx,y,Δf⁎/∑x,yIx,y,Δf]. (4) If we determine the search direction ( dx,y(j)), then the feedback parameter α is determined by solving the following equation: ∂E(j)(ϕx,y(j)−α·dx,y(j))∂α=0. (5) Although the MAL method can retrieve low-frequency components of the electron phase more precisely via the iterative processes, the method requires a much longer processing time than the Fourier analytical reconstruction methods. In this paper, we developed a software code that utilizes a graphical processing unit (GPU) in order to shorten the processing time of the MAL method. As a result, the processing time of the method was accelerated to ~70 times faster than the processing time with a central processing unit. Materials and methods We observed magnetic bubbles generated in the single-crystal thin films of BaFe10.35Sc1.6Mg0.05O19 [24]. The magnetic bubbles were generated by applying a magnetic field (~200 mT) perpendicular to the sample using an objective lens of a transmission electron microscope (JEOL JEM-2100F, 200 kV). We observed type-I and type-II magnetic bubbles [25] in the thin films. Type-I magnetic bubbles are generally characterized by magnetic domain walls rotationally magnetized in either a clockwise or a counterclockwise direction. Type-II magnetic bubbles are generally characterized by two parallel domain walls with a pair of Bloch lines. The 32 through-focus images of the magnetic bubbles were obtained in a defocus range from −3.36 mm to 3.15 mm with a 210-μm focus interval using an objective mini-lens of the TEM. In this study, the positive value of the defocus axis is set as the under-focus side. The captured through-focus images were aligned using a template-matching method [26] to correct changes in optical parameters such as image rotation, image magnification and image shift. In this scheme, the distance and positional relations between the same types of magnetic bubbles were measured using the template-matching method for each defocused image. The through-focus images were then aligned so that the distance and positional relationship of the bubbles were kept constant between each defocused image. Results and discussion Simulation We evaluated the CG-MAL method using simulated through-focus images of a type-I magnetic bubble. Figure 1a and b shows an assumed phase of electron wave function induced by the magnetic bubble and its line profile along the radial direction, respectively. In this study, the phase shift was assumed to be rotationally symmetric. The maximum phase shift was assumed to be 90° at the center of the magnetic bubble. The phase changes to zero in the form of a half-cycle of a cosine function at the edges of the magnetic bubble. Figure 1c shows the corresponding in-plane magnetization map of the magnetic bubble. In Fig. 1c, the color and brightness correspond to the direction and magnitude of the in-plane magnetization as shown in the color wheel. The in-plane magnetization of the magnetic domain wall was assumed to be localized in the ring-shaped region as indicated by white-colored arrows. The magnetic field in the center of the magnetic bubble is assumed to be upward along the optical axis. By contrast, the magnetic field in the outskirt dark regions is downward along the optical axis. Fig. 1. View largeDownload slide (a) Assumed phase of electron wave function and (b) its line profile along the radial direction of the type-I magnetic bubble. (c) The corresponding in-plane magnetization map of a type-I magnetic bubble; the in-plane magnetization vector is represented by white-colored arrows. The color and brightness in the in-plane magnetization map correspond to the direction and magnitude of the in-plane magnetization, respectively, as shown in the color wheel. Fig. 1. View largeDownload slide (a) Assumed phase of electron wave function and (b) its line profile along the radial direction of the type-I magnetic bubble. (c) The corresponding in-plane magnetization map of a type-I magnetic bubble; the in-plane magnetization vector is represented by white-colored arrows. The color and brightness in the in-plane magnetization map correspond to the direction and magnitude of the in-plane magnetization, respectively, as shown in the color wheel. Figure 2 shows a few LTEM images selected from the 32 simulated through-focus images. These images were calculated under the same defocus conditions as those used in the LTEM experiment. With respect to the in-focus condition, the contrast of the magnetic bubble disappeared, as shown in Fig. 2a, because the type-I magnetic bubble was assumed to be an ideal phase object in the calculation. Figure 2b–e depicts the LTEM images corresponding to the defocus values of ±0.4 mm and ±3.2 mm. By performing large-scale defocusing, we obtained strong contrast in the images of the magnetic bubble, as shown in Fig. 2c and e. The ring-shaped bright and dark contrasts in the LTEM images are understood to be formed by the convergence and divergence of the incident electron beam, respectively. Fig. 2. View largeDownload slide Simulated LTEM images of the type-I magnetic bubble at defocus values corresponding to (a) 0.0 mm, (b) 0.4 mm, (c) 3.2 mm, (d) −0.4 mm and (e) −3.2 mm. The positive value of the defocus axis is set as the under-focus side. Fig. 2. View largeDownload slide Simulated LTEM images of the type-I magnetic bubble at defocus values corresponding to (a) 0.0 mm, (b) 0.4 mm, (c) 3.2 mm, (d) −0.4 mm and (e) −3.2 mm. The positive value of the defocus axis is set as the under-focus side. We performed the phase retrieval using the 32 through-focus images, including the images shown in Fig. 2. Figure 3a shows the phase of the wave function reconstructed by the 3DFFM used as the initial wave function for the CG-MAL method. In Fig. 3a, the low-spatial-frequency components of the phase distribution were not well retrieved because of the effective contrast transfer function [27] of the Fourier analytical reconstruction method. Figure 3b shows the corresponding in-plane magnetization map obtained by differentiating the retrieved phase distribution [28]. The in-plane magnetization map shows several ring artifacts caused by the sub-peaks in the obtained phase distribution. The simulation results indicate that it is not possible to reconstruct the whole wave function under the current experimental conditions by only using such Fourier analytical reconstruction methods. Fig. 3. View largeDownload slide (a), (b) Retrieved phase distribution and corresponding magnetization map obtained using the 3DFFM. The phase distribution and magnetization map obtained by the CG-MAL method after (c), (d) 10 iterations and after (e), (f) 100 iterations. (g) The line profile of the retrieved phase distribution in (e) along the radial direction. Fig. 3. View largeDownload slide (a), (b) Retrieved phase distribution and corresponding magnetization map obtained using the 3DFFM. The phase distribution and magnetization map obtained by the CG-MAL method after (c), (d) 10 iterations and after (e), (f) 100 iterations. (g) The line profile of the retrieved phase distribution in (e) along the radial direction. Figure 3c–f depicts the reconstructed phase distributions and the corresponding in-plane magnetization maps obtained by the CG-MAL method after 10 and 100 iterations, respectively. By repeating the iterative procedures, we could appropriately enhance the low-spatial-frequency components and obtain an in-plane magnetization map equivalent to that assumed in Fig. 1c. Figure 3g shows the line profile of the retrieved phase distribution in Fig. 3e; it demonstrates that the accuracy of the phase retrieval is substantially high compared with that shown in Fig. 1b. In Fig. 3g, the retrieved phase distribution shifts by ~10° in the negative direction because it is theoretically impossible to obtain the offset value of the phase distribution in the focal-series reconstruction methods. Thus, we confirm that the phase distribution of the electron wave is theoretically retrieved by the CG-MAL method. In the CG-MAL method, the detection limit of the phase angle mainly depends on the total number of defocused images, defocus range of the through-focus images, size of the magnetic bubbles and the electron dose. If these parameters are determined, we can estimate the phase-angle determination precision on the basis of the image simulation by considering the electron dose. We have confirmed that the detection limit of the method is approximately a few degrees under the current simulation parameters and the experimental electron dose density (50 electrons/(nm2·frame)). Experimental results We applied the CG-MAL method to the experimentally obtained through-focus images. Figure 4 shows several LTEM images selected from the experimentally obtained 32 through-focus images of the magnetic bubbles. At the in-focus position, the contrast of magnetic bubbles almost disappears, as shown in Fig. 4a. This lack of contrast indicates that the magnetic domain structures of the sample are considered as phase objects. In Fig. 4b and d, we observe a slight contrast of the magnetic bubbles at the defocus values of ±0.4 mm although the contrast is relatively weak. The magnetic bubbles indicated by A and C in Fig. 4b exhibit contrast typical of the type-I magnetic bubbles. The magnetic bubbles indicated by B and D in Fig. 4b exhibit contrast typical of the type-II magnetic bubbles. Conversely, in Fig. 4c and e, we observe stronger Fresnel contrast of the magnetic bubbles compared with those in Fig. 4b and d. Fig. 4. View largeDownload slide Experimentally obtained through-focus images of the magnetic bubbles at defocus values corresponding to (a) 0.0 mm, (b) 0.4 mm, (c) 3.2 mm, (d) −0.4 mm and (e) −3.2 mm. The magnetic bubbles indicated by A and C show the typical contrast of type-I magnetic bubbles. The magnetic bubbles indicated by B and D show the typical contrast of type- II magnetic bubbles. Fig. 4. View largeDownload slide Experimentally obtained through-focus images of the magnetic bubbles at defocus values corresponding to (a) 0.0 mm, (b) 0.4 mm, (c) 3.2 mm, (d) −0.4 mm and (e) −3.2 mm. The magnetic bubbles indicated by A and C show the typical contrast of type-I magnetic bubbles. The magnetic bubbles indicated by B and D show the typical contrast of type- II magnetic bubbles. Figure 5a and b shows the phase distribution and corresponding in-plane magnetization map obtained by the TIE method, respectively. In the TIE method, the retrieved phase and in-plane magnetization vary depending on the three applied defocused images. In the present work, we selected Fig. 4a, b and d as the three images based on the defocus value used in the previously reported TIE method [20]. To eliminate image artifacts, the phase distribution obtained by the TIE method was processed by a low-cut filter that eliminates low-spatial-frequency components with frequencies <2.5 μm−1. The phase distribution and magnetization map appear as relatively well reconstructed, although the magnetization map includes a few noise components and artifacts caused by the bend contour lines. Fig. 5. View largeDownload slide Retrieved phase distribution and corresponding in-plane magnetization map obtained by (a), (b) the TIE method and by (c), (d) the CG-MAL method with 100 iterations. Results (a) and (b) corresponding to the TIE method were obtained using the three LTEM images in Fig. 4(a), (b) and (d). Results (c) and (d) corresponding to the CG-MAL method were obtained by using 32 through-focus images. Fig. 5. View largeDownload slide Retrieved phase distribution and corresponding in-plane magnetization map obtained by (a), (b) the TIE method and by (c), (d) the CG-MAL method with 100 iterations. Results (a) and (b) corresponding to the TIE method were obtained using the three LTEM images in Fig. 4(a), (b) and (d). Results (c) and (d) corresponding to the CG-MAL method were obtained by using 32 through-focus images. Figure 5c and d shows the phase distribution and corresponding in-plane magnetization maps obtained by the CG-MAL method with 100 iterations. The phase distribution and the magnetization map are qualitatively the same as those obtained using the TIE method. Therefore, we experimentally retrieved the phase distribution of the electron wave by the CG-MAL method by using several through-focus images obtained under the defocus range from approximately −3 mm to 3 mm. The current defocus range is several tens of times larger than the previous range studied using the TIE method [20, 21, 24]. Because the CG-MAL method can treat numerous through-focus images in a wide defocus range, the obtained results appear to be more reliable and precise, reducing the noise and artifacts. We believe that the CG-MAL method is superior to the TIE method in this regard, although the CG-MAL method requires a much longer processing time. In Fig. 5d obtained by the CG-MAL method, we observed magnetic domain structures of the magnetic bubbles in detail. For example, the type-I magnetic bubbles indicated by A and C include magnetic domain walls localized in a region with a width of ~109 nm. In addition, the type-II magnetic bubbles indicated by B and D include slightly different detailed magnetic domain structures. In the magnetic bubble indicated by B, both ends of two parallel domain walls are shifted relative to each other. By contrast, both ends of two parallel domain walls face each other in the magnetic bubble indicated by D. In addition, the magnitude of in-plane magnetization is different in the center region of each magnetic bubble. These differences can be caused by differences in the 3D structure of each magnetic bubble. The use of micro-magnetic simulations [29] and the obtained in-plane magnetization map will enable a quantitative discussion of the difference in each 3D magnetic domain structure in the future. Comparison of convergence behavior In this section, we compare the convergence behavior of the CG-MAL method to those of the convergence behaviors of the SD-MAL method and the IWFR method, which correspond to other popular iterative phase retrieval methods used in HRTEM. The IWFR method is based on a Gerchberg–Saxton type algorithm [30] in which a wave function is reconstructed by an iterative 2D Fourier transform and an inverse 2D Fourier transform. These iterative reconstruction methods can also reconstruct the same phase distributions as the CG-MAL method. Figure 6a and b shows the change in the error functions of the IWFR, SD-MAL and CG-MAL methods for the simulation and experiment, as described in Sections 4.1 and 4.2, respectively. Each value of the error functions was calculated based on Eq. (2). The same initial wave function is used for each phase retrieval method; thus, the values of the error functions of the first iteration are identical to each other. In Fig. 6b, the error function of each iteration method does not converge to zero because the experimentally obtained through-focus images include random noise components. The convergence speed of the CG-MAL method exceeds that of the other methods in both the simulation and the experiment. Therefore, we conclude that the CG-MAL method exhibits better convergence behavior than the other reconstruction methods. In addition, the IWFR and SD-MAL methods exhibit similar convergence behaviors. One reason for these similar behaviors is that the Gerchberg–Saxton type algorithm is mathematically equivalent to the steepest descent method [31]. Fig. 6. View largeDownload slide Change in the error functions of the IWFR, SD-MAL and CG-MAL methods for (a) the simulation and (b) the experiment. Fig. 6. View largeDownload slide Change in the error functions of the IWFR, SD-MAL and CG-MAL methods for (a) the simulation and (b) the experiment. In the present study, the processing time of the CG-MAL and SD-MAL methods was ~77 ms per iteration using a GPU (NVIDIA GTX 1070). By contrast, the processing time of the IWFR method using the GPU was ~35 ms. The processing time per iteration of the CG-MAL method is slightly longer than that of the IWFR method ; however, the CG-MAL method enables the converged data to be obtained in much shorter time. In the present study, the image size of the 32 through-focus images corresponds to 1024 × 1024 px, which includes the original image region (512 × 512 px) and the padding regions surrounding the original images with a width of 256 px [18]. Concluding remarks In this study, we reported phase retrieval for magnetic domain structures via the CG-MAL method, which is usually applied to HRTEM observations. As an application example, we evaluated the method using 32 simulated and experimentally obtained through-focus images of magnetic bubbles where the images were collected under a defocus range from approximately −3 mm to 3 mm. As a result, we successfully obtained the magnetic domain structures of the magnetic bubbles. The current defocus range is several tens of times higher than the previous range examined using the TIE method. Furthermore, the CG-MAL method exhibited better convergence behavior than the other iterative reconstruction methods. The method will also be widely and effectively applicable to the observation of magnetic domain structures other than magnetic bubbles. Therefore, the CG-MAL method is promising for the study of detailed magnetic domain structures when several highly defocused through-focus images are used. In the experimental conditions used in the present study, the maximum defocus range of the through-focus images was mainly restricted by the correction accuracy of the optical parameters such as image rotation, image magnification and image shift. Thus, performing further detailed analyses of magnetic domain structures necessitates the development of a TEM system that enables large-scale defocusing with smaller changes in the optical parameters. In the near future, further quantitative observations of magnetic domain structures will be possible through the combination of the TEM system and the phase retrieval method. Funding Japan Society for the Promotion of Science Grant-in-Aid for Scientific Research (B) (grant number 25286059). References 1 Nakajima H , Kotani A , Harada K , Ishii Y , and Mori S ( 2016 ) Foucault imaging and small-angle electron diffraction in controlled external magnetic fields . Microscopy 65 : 473 – 478 . Google Scholar CrossRef Search ADS PubMed 2 Saxton W O ( 1994 ) What is the focus variation method? Is it new? Is it direct? Ultramicroscopy 55 : 171 – 181 . Google Scholar CrossRef Search ADS 3 Shibata N , Findlay S D , Kohno Y , Sawada H , Kondo Y , and Ikuhara Y ( 2012 ) Differential phase-contrast microscopy at atomic resolution . Nat. Phys. 8 : 611 – 615 . Google Scholar CrossRef Search ADS 4 McGrouther D , Lamb R J , Krajnak M , McFadzean S , McVitie S , Stamps R L , Leonov A O , Bogdanov A N , and Togawa Y ( 2016 ) Internal structure of hexagonal skyrmion lattices in cubic helimagnets . New J. Phys. 18 : 095004 . Google Scholar CrossRef Search ADS 5 Harada K , Matsuda T , Tonomura A , Akashi T , and Togawa Y ( 2006 ) Triple-biprism electron interferometry . J. Appl. Phys. 99 : 113502 . Google Scholar CrossRef Search ADS 6 Niermann T , and Lehmann M ( 2016 ) Holographic focal series: differences between inline and off-axis electron holography at atomic resolution . J. Phys. D: Appl. Phys. 49 : 194002 . Google Scholar CrossRef Search ADS 7 Teague M R ( 1983 ) Deterministic phase retrieval: a Green’s function solution . J. Opt. Soc. Am. 73 : 1434 – 1441 . Google Scholar CrossRef Search ADS 8 Paganin D , and Nugent K A ( 1998 ) Noninterferometric phase imaging with partially coherent light . Phys. Rev. Lett. 80 : 2586 – 2589 . Google Scholar CrossRef Search ADS 9 Ishizuka K , and Allman B ( 2005 ) Phase measurement of atomic resolution image using transport of intensity equation . J. Electron Microsc. 54 : 191 – 197 . 10 Mitome M , Ishizuka K , and Bando Y ( 2010 ) Quantitativeness of phase measurement by transport of intensity equation . J. Electron Microsc. 59 : 33 – 41 . Google Scholar CrossRef Search ADS 11 Kawasaki T , Takai Y , Ikuta T , and Shimizu R ( 2001 ) Wave field restoration using three-dimensional fourier filtering method . Ultramicroscopy 90 : 47 – 59 . Google Scholar CrossRef Search ADS PubMed 12 Nomaguchi T , Kimura Y , and Takai Y ( 2008 ) High-resolution wave field reconstruction using a through-focus series of images taken under limited electron dose . Ultramicroscopy 108 : 1520 – 1528 . Google Scholar CrossRef Search ADS PubMed 13 Dyck D V , Beeck M O , and Coene W ( 1993 ) A new approach to object wave-function reconstruction in electron microscopy . Optik 93 : 103 – 107 . 14 Utsuro H , Ando T , Takai Y , Ikuta T , and Shimizu R ( 2001 ) Theoretical background of defocus image modulation processing (DIMP) based on three dimensional optical transfer functions (3D-OTFs) . Optik 112 : 67 – 75 . Google Scholar CrossRef Search ADS 15 Tamura T , Kimura Y , and Takai Y ( 2017 ) Development of a real-time wave field reconstruction TEM system (I): incorporation of an auto focus tracking system . Microscopy 66 : 172 – 181 . Google Scholar CrossRef Search ADS PubMed 16 Allen L J , McBride W , O’Leary N L , and Oxley M P ( 2004 ) Exit wave reconstruction at atomic resolution . Ultramicroscopy 100 : 91 – 104 . Google Scholar CrossRef Search ADS PubMed 17 Coene W M J , Thust A , Beeck M O , and Dyck D V ( 1996 ) Maximum-likelihood method for focus-variation image reconstruction in high resolution transmission electron microscopy . Ultramicroscopy 64 : 109 – 135 . Google Scholar CrossRef Search ADS 18 Lin F , Chen F R , Chen Q , Tang D , and Peng L M ( 2006 ) The wrap-around problem and optimal padding in the exit wave reconstruction using HRTEM images . J. Electron Microsc. 55 : 191 – 200 . Google Scholar CrossRef Search ADS 19 Ophus C , and Ewalds T ( 2012 ) Guidelines for quantitative reconstruction of complex exit waves in HRTEM . Ultramicroscopy 113 : 88 – 95 . Google Scholar CrossRef Search ADS 20 Kotani A , Nakajima H , Harada K , Ishii Y , and Mori S ( 2016 ) Lorentz microscopy and small-angle electron diffraction study of magnetic textures in La1−xSrxMnO3 (0.15<x<0.30): the role of magnetic anisotropy . Phys. Rev. B 94 : 024407/1 – 024407/7 . Google Scholar CrossRef Search ADS 21 Nakajima H , Kawase H , Kurushima K , Kotani A , Kimura T , and Mori S ( 2017 ) Observation of magnetic domain and bubble structures in magnetoelectric Sr3Co2Fe24O41 . Phys. Rev. B 96 : 024431 . Google Scholar CrossRef Search ADS 22 Kimoto K , Sawada H , Sasaki T , Sato Y , Nagai T , Ohwada M , Suenaga K , and Ishizuka K ( 2013 ) Quantitative evaluation of temporal partial coherence using 3D Fourier transforms of through-focus TEM images . Ultramicroscopy 134 : 86 – 93 . Google Scholar CrossRef Search ADS PubMed 23 Kawasaki T , Utsuro H , Takai Y , and Shimizu R ( 1999 ) Evaluation of image drift correction by three-dimensional Fourier analysis . J. Electron Microsc. 48 : 35 – 37 . Google Scholar CrossRef Search ADS 24 Nakajima H , Kotani A , Harada K , Ishii Y , and Mori S ( 2016 ) Formation mechanisms of magnetic bubbles in an M-type hexaferrite: role of chirality reversal at domain walls . Phys. Rev. B 94 : 224427 . Google Scholar CrossRef Search ADS 25 Suzuki T ( 1983 ) A study of magnetization distribution of submicron bubbles in sputtered Ho-Co thin films . J. Magn. Magn. Mater. 31-34 : 1009 – 1010 . Google Scholar CrossRef Search ADS 26 Lewis J P ( 1995 ) Fast template matching . Vis. Interface 15-19 : 120 – 123 . 27 Ando T , Taniguchi Y , Takai Y , Kimura Y , Shimizu R , and Ikuta T ( 1994 ) Development of real-time defocus-modulation-type active image processing (DMAIP) for spherical-aberration-free TEM observation . Ultramicroscopy 54 : 261 – 267 . Google Scholar CrossRef Search ADS 28 Lichte H , and Lehmann M ( 2008 ) Electron holography—basics and applications . Rep. Prog. Phys. 71 : 016102 . Google Scholar CrossRef Search ADS 29 Lopez-Diaz L , Aurelio D , Torres L , Martinez E , Hernandez-Lopez M A , Gomez J , Alejos O , Carpentieri M , Finocchio G , and Consolo G ( 2012 ) Micromagnetic simulations using graphics processing units . J. Phys. D: Appl. Phys. 45 : 323001 . Google Scholar CrossRef Search ADS 30 Gerchberg R W , and Saxton W O ( 1972 ) A practical algorithm for the determination of the phase from image and diffraction plane pictures . Optik 35 : 237 . 31 Fienup J R ( 1982 ) Phase retrieval algorithms: a comparison . Appl. Opt. 21 : 2758 – 2769 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of The Japanese Society of Microscopy. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Microscopy – Oxford University Press

**Published: ** Mar 26, 2018

Loading...

personal research library

It’s your single place to instantly

**discover** and **read** the research

that matters to you.

Enjoy **affordable access** to

over 18 million articles from more than

**15,000 peer-reviewed journals**.

All for just $49/month

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Read from thousands of the leading scholarly journals from *SpringerNature*, *Elsevier*, *Wiley-Blackwell*, *Oxford University Press* and more.

All the latest content is available, no embargo periods.

## “Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”

Daniel C.

## “Whoa! It’s like Spotify but for academic articles.”

@Phil_Robichaud

## “I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”

@deepthiw

## “My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”

@JoseServera

DeepDyve ## Freelancer | DeepDyve ## Pro | |
---|---|---|

Price | FREE | $49/month |

Save searches from | ||

Create folders to | ||

Export folders, citations | ||

Read DeepDyve articles | Abstract access only | Unlimited access to over |

20 pages / month | ||

PDF Discount | 20% off | |

Read and print from thousands of top scholarly journals.

System error. Please try again!

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

Copy and paste the desired citation format or use the link below to download a file formatted for EndNote

**EndNote**

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.

ok to continue