Bayesian object classification in nanoimages
In this mini-dissertation the importance of having an automated object classification procedure for classifying nanoparticles in nanoscale images (or referred to as nanoimages in this mini-dissertation) is discussed, and a detailed overview of such a procedure, proposed by Konomi et al. (2013) is pr...
| Main Author: | |
|---|---|
| Other Authors: | |
| Language: | en |
| Published: |
University of Pretoria
2018
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/2263/63790 Haywood, AS 2017, Bayesian object classification in nanoimages, MSc Mini Dissertation, University of Pretoria, Pretoria, viewed yymmdd <http://hdl.handle.net/2263/63790> |
| Summary: | In this mini-dissertation the importance of having an automated object classification procedure for classifying nanoparticles in nanoscale images (or referred to as nanoimages in this mini-dissertation) is discussed, and a detailed overview of such a procedure, proposed by Konomi et al. (2013) is provided, with emphasis on applying the procedure to nanoimages of gold nanoparticles. In the process a simplified approach to classifying occluded objects when dealing with homogeneously shaped objects is introduced. Nanotechnology is a technology that deals with measurements obtained in nano-scale (one billionth of a metre), and for ease of reference these images will henceforth be referred to as nanoimages. The focus is restricted to nanoimages, obtained using a Transmission Electron Microscope (TEM). A common phenomenon that occurs during the image capturing is occlusion of objects in the image. This occlusion leads to some unwanted results during the image analysis phase, making the use of a more sophisticated classification algorithm necessary. An automated classification algorithm that successfully deals with occluded objects in nanoimages is discussed and a detailed discussion on the implementation of this algorithm is provided. The techniques used in the algorithm involve a combination of several Bayesian techniques to classify the objects in the nanoimage. Markov Chain Monte Carlo (MCMC) sampling techniques are used to simulate the unknown posterior, with samplers ranging from the Metropolis-Hastings and Reversable Jumps MCMC samplers to Monte Carlo Metropolis Hastings samplers used in obtaining the simulated posterior. Since one of the main objectives of this investigation will be the processing of images, a discussion on the most widely used image processing techniques is provided, with specific focus on how these techniques are used to extract objects of interest from the image. An overview of nanotechnology and its applications is provided, along with a variability study for the capturing of nanoimages using TEM. The aim of the study is to introduce controlled variability in the sampling through imposing specific sampling conditions, in order to determine if imposing these conditions significantly affects the measurements obtained. This variability study, according to our knowledge, is the first performed at this level of detail, and provides very useful considerations when performing a nanoimage study. === Mini Dissertaion (MSc)--University of Pretoria, 2017. === NRF (under CSUR grant 90315) === CSIR === Statistics === MSc === Unrestricted |
|---|