Bayesian Models for Image Object Classification with Application to Gold Nanoparticles

• Activity Number: 345
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Bledar Konomi*+
• Companies: Texas A & M University
• Keywords: Object classification ; Statistical shape analysis ; Nanoparticles ; Markov-chain Monte-carlo ; RJ-MCMC ; Monte Carlo Metropolis Hasting

Abstract

By controlling the shape and size of nanoparticles during synthesis, one could control the properties of the synthesized material. Transmission electron microscopy imaging technique can be used to measure the morphological characteristics of nanoparticles, which can be simple circles or more complex irregular polygons with varying degree of scales and sizes. A major difficulty in analyzing the TEM images is the overlapping of objects having different morphological properties with no specific information about the number of objects present. Also the objects lying along the boundary render automated image analysis much more difficult. To overcome these challenges, we propose a Bayesian image segmentation and classification procedure based on the marked-point process representation of the objects. We derive models, both for the marks which parametrize the morphological aspects and the points which determine the location of the objects, to greatly reduce the complexity of the problem. We carry-out the inference by using a novel Markov chain Monte Carlo (MCMC) technique since the posterior distribution is analytically intractable.