Nanoparticle Localization Using Gabor Filters

Abstract

Gabor filters are described as functions product of a Gaussian-shaped kernel and a complex sinusoid function that minimizes the uncertainty principle between the spatial and frequency domains [1, 2]. These filters are specially utilized in computer vision for texture analysis and feature extraction of images, due the nature of the function itself it is possible to extract observations that are well defined and repetitive in the 2D space [3]. For this reason, the high periodicity registered in high resolution transmission electron microscopy (HRTEM) images of crystalline nanoparticles as well as and their shape morphology is highly exploitable. In this work we present a comprehensive method based on Gabor filters to localize in an automatic and efficient manner the nanoparticles, which can be implemented in dynamic studies using in situ transmission electron microscopy (TEM). In this way, orientation, and frequency of appearance of the features observed in the input signal are relevant to analyze the specimen, for instance, planes are rotated around the space occupied by each particle. An important parameter to be considered is the wavelength of the passed filters, which may be determined by the interplanar spacings of the high-resolution images.