Date of Award
Program or Major
Electrical and Computer Engineering
Master of Science
Richard A Messner
Michael J Carter
Wayne J Smith
While significant strides in neural network and machine vision applications have been made in recent years, humans still remain the most proficient at feature extraction and pattern recognition tasks. Some researchers have attempted to utilize select aspects of the human visual system in order to perform application-specific visual tasks. However, none have been able to develop a computational model of the biological human visual system that can perform the many complex pattern recognition tasks that we do as humans. This thesis focuses on significant improvements to an existing human visual system model created by N. Radhi, and the novel implementation of a deep learning system for road detection utilizing non-uniformly sampled images in log-polar coordinate space. A convolutional neural network is used to compare the non-uniformly sampled image model with the conventional uniform structure, with the non-uniform model demonstrating significant increases in processing speed while retaining high validation accuracy. Comparisons between the uniform and non-uniform models when subjected to a variety of preprocessing methods are presented.
Koester, Evan, "A Human Visual System Inspired Feature Recognition Method Using Convolutional Neural Networks" (2018). Master's Theses and Capstones. 1231.