Image analysis for measuring rod network properties

Abstract

In recent years, metallic nanowires have been attracting significant attention as next-generation flexible transparent conductive films. The performance of films depends on the network structure created by nanowires. Gaining an understanding of their structure, such as connectivity, coverage, and alignment of nanowires, requires the knowledge of individual nanowires inside the microscopic images taken from the film. Although nanowires are flexible up to a certain extent, they are usually depicted as rigid rods in many analysis and computational studies. Herein, we propose a simple and straightforward algorithm based on the filtering in the frequency domain for detecting the rod-shape objects inside binary images. The proposed algorithm uses a specially designed filter in the frequency domain to detect image segments, namely, the connected components aligned in a certain direction. Those components are post-processed to be combined under a given merging rule in a single rod object. In this study, the microscopic properties of the rod networks relevant to the analysis of nanowire networks were measured for investigating the opto-electric performance of transparent conductive films and their alignment distribution, length distribution, and area fraction. To verify and find the optimum parameters for the proposed algorithm, numerical experiments were performed on synthetic images with predefined properties. By selecting proper parameters, the algorithm was used to investigate silver nanowire transparent conductive films fabricated by the dip coating method.

Neural Network for Nanoscience Scanning Electron Microscope Image Recognition

“…In particular, 1D nanostructures such as nanowires are used in a diverse range of applications from photonics to energy devices 17 , 18 . Factors such as the length, diameter, number density and orientation of the nanowires can determine the performance of such devices 19 , 20 .…”

“…Thus, alternative image processing methods have been developed 21 , 22 . For example, the Fast Fourier Transform method has been used to identify the periodic behaviour in an image 20 , 23 . The local gradient method, which is a particular case of regression method, has been used to quantify orientation and isotropy properties of a region of interest in an image 24 , based on the evaluation of the structure tensor in a local neighborhood 25 .…”

Image Segmentation for FIB-SEM Serial Sectioning of a Si/C–Graphite Composite Anode Microstructure Based on Preprocessing and Global Thresholding

“…Figure 5 compares the performance of curtain artifacts reduction by using the MGNRF, the Gaussian/notch reject filter (GNRF) (Kim et al, 2015), and the INRF, where Figures 5b to 5d show the corresponding filters and the resultant filtering applied to Figure 5a under various cutoff frequencies, respectively. For the INRF, the pass-band width (length of the white horizontal region in the center of the filter) is 2 D 0,GL , and the reject-band width of INRF and GNRF (the width of the black bar) is 2 D 0,GK .…”

“…Typically, the curtain artifacts are aligned in the y -direction (beam direction), like the vertical line patterns marked by the arrows in Figure 3a. When a given image is displayed in a frequency domain via Fourier transform, these line patterns can be merged into a single bright horizontal band as shown in Figure 3b, which is perpendicular to the direction of the patterns in the original image (Kim et al, 2015). Given that the materials inside the ROI do not have distinct vertical features, the removal of the horizontal band does not spoil the features of the materials significantly.…”

“…Given that the materials inside the ROI do not have distinct vertical features, the removal of the horizontal band does not spoil the features of the materials significantly. In this study, we designed a modified Gaussian/notch reject filter (MGNRF), which is a modified version of our previous designed filter to extract linear features (Kim et al, 2015). The proposed filter is modified to reduce ringing artifacts that occur when a filter has an abrupt change in magnitude such as an ideal filter (Gonzalez & Woods, 2008).…”