Type of presentation: Poster

IT-16-P-2686 Remove the CCD influence from high-resolution electron microscopy images

Lin F.¹, Jin C.², Yang Y.¹

¹College of Science, South China Agricultural University, Guangzhou, Guangdong 510642, PR China, ²State Key Laboratory of Silicon Materials, Department of Materials Science and Engineering, Zhejiang University, Hangzhou, Zhejiang 310027, PR China

flin_163@163.com

Slow-scan charge-coupled device (CCD) camera is now an important recording medium for high-resolution transmission electron microscopy (HRTEM). However, the pixel of CCD chips has a certain size, and then the point spread effect cannot be neglected in image. In frequency domain, signal transfer is described by a modulation transfer function (MTF). In addition, Possion noise is inherent in electron images. The noise transfer is described by noise transfer function (NTF), which gives the attenuation of noise power relying on its frequencies [1].
To remove the MTF from HRTEM image, the MTF should be calculated firstly from experiments. All experimental images were recorded on a JEM-2010F TEM equipped with a Gatan-894 CCD camera. The beam-stopper image is shown in Fig. 1(a). Through averaging the image intensities of positions equally distancing from the edge, we could get the edge profile shown in Fig. 2(b). Deconvoluted from the filtered profile of edge, the PSF of CCD was resolved. We show it in Fig. 1(c) and use the Fourier transform to get the MTF of CCD.
To measure the NTF, 16 uniform-illumination images without beam-stoppers and 8 dark-current images were recorded. The ‘NTF 1’ and ‘NTF 2’ in Fig. 2 are estimated from the 16 uniform illumination images of totally ~2190 and ~4760 counts, respectively. Although the electron doses are different, the noise characteristics of NTF are almost the same. The NTF is mainly affected by the accelerated voltage and exposure time.
Based on an improved Wiener deconvolution filter, the restored image I’(u,v) in frequency domain of (u,v) is calculated as [2],
I'(u,v)=I(u,v)MTF^*(u,v)/{|MTF(u,v)|²+P_n(u,v)/[P_D(u,v)-P_n(u,v)]},
in which, I(u,v) is a HRTEM image, MTF(u,v) is the MTF of CCD in 2-dimensional space, and P_D(u,v) and P_n(u,v) are the mean power spectral densities of detected image I(u, v) and additive noise, respectively. P_D(u,v) is actually the image diffraction pattern multiplied by its conjugate, and P_n(u,v) is estimated from NTF. The NTF provides the power distributions of noise at various frequencies and is measured from uniform-illumination images. For weak scattering objects, such as few-layer graphene or boron nitride, the image is considerably “uniform” because of the weak scattering of atoms. Fig 3(a) gives a raw HRTEM image of graphene. After deconvolution, the lattices in center region are resolved from noise.

Authors acknowledged supports from the National Science Foundation of China (61172011 and 51222202), National Basic Research Program of China (2014CB932500), the Program for Innovative Research Team in University of Ministry of Education of China (IRT13037), the Fundamental Research Funds for the Central Universities (2014XZZX003-07) and Guangdong Natural Science Foundation (10151064201000006).

Fig. 1: (a) The beam-stopper image. The white line indicates one of the beam-stopper edges. (b) The blue line is the edge profiles averaged from one edge in (a). The red line gives a filtered edge profile. (c) The CCD’s PSF calculated from (b). Ideal step function convoluting with this PSF is the red line in (b).	Fig. 2: The NTF of CCD. ‘NTF 1’ and ‘NTF 2’ are calculated from the images uniformly illuminated under different electron doses after normalization.
Fig. 3: (a) A raw HRTEM image of graphene. (b) The image restored with the improved Wiener deconvolution filter.