Login | Register

A gaussian mixture-based approach to synthesizing nonlinear feature functions for automated object detection

Title:

A gaussian mixture-based approach to synthesizing nonlinear feature functions for automated object detection

Guo, Pei Fang (2010) A gaussian mixture-based approach to synthesizing nonlinear feature functions for automated object detection. PhD thesis, Concordia University.

[thumbnail of NR67351.pdf]
Preview
Text (application/pdf)
NR67351.pdf - Accepted Version
3MB

Abstract

Feature design is an important part to identify objects of interest into a known number of categories or classes in object detection. Based on the depth-first search for higher order feature functions, the technique of automated feature synthesis is generally considered to be a process of creating more effective features from raw feature data during the run of the algorithms. This dynamic synthesis of nonlinear feature functions is a challenging problem in object detection. This thesis presents a combinatorial approach of genetic programming and the expectation maximization algorithm (GP-EM) to synthesize nonlinear feature functions automatically in order to solve the given tasks of object detection. The EM algorithm investigates the use of Gaussian mixture which is able to model the behaviour of the training samples during an optimal GP search strategy. Based on the Gaussian probability assumption, the GP-EM method is capable of performing simultaneously dynamic feature synthesis and model-based generalization. The EM part of the approach leads to the application of the maximum likelihood (ML) operation that provides protection against inter-cluster data separation and thus exhibits improved convergence. Additionally, with the GP-EM method, an innovative technique, called the histogram region of interest by thresholds (HROIBT), is introduced for diagnosing protein conformation defects (PCD) from microscopic imagery. The experimental results show that the proposed approach improves the detection accuracy and efficiency of pattern object discovery, as compared to single GP-based feature synthesis methods and also a number of other object detection systems. The GP-EM method projects the hyperspace of the raw data onto lower-dimensional spaces efficiently, resulting in faster computational classification processes.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (PhD)
Authors:Guo, Pei Fang
Pagination:xiii, 79 leaves : ill. ; 29 cm.
Institution:Concordia University
Degree Name:Ph. D.
Program:Electrical and Computer Engineering
Date:2010
Thesis Supervisor(s):Bhattacharya, P and Kharma, N
Identification Number:LE 3 C66E44P 2010 G86
ID Code:979537
Deposited By: Concordia University Library
Deposited On:09 Dec 2014 18:01
Last Modified:13 Jul 2020 20:12
Related URLs:
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top