Articles via Databases
Articles via Journals
Online Catalog
E-books
Research & Information Literacy
Interlibrary loan
Theses & Dissertations
Collections
Policies
Services
About / Contact Us
Administration
Littman Architecture Library
This site will be removed in January 2019, please change your bookmarks.
This page will redirect to https://digitalcommons.njit.edu/dissertations/1383/ in 5 seconds

The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title: Novel image descriptors and learning methods for image classification applications
Author: Puthenputhussery, Ajit
View Online: njit-etd2018-040
(xiv, 110 pages ~ 29.1 MB pdf)
Department: Department of Computer Science
Degree: Doctor of Philosophy
Program: Computer Science
Document Type: Dissertation
Advisory Committee: Liu, Chengjun (Committee chair)
Mili, Ali (Committee member)
Narahara, Taro (Committee member)
Oria, Vincent (Committee member)
Basu Roy, Senjuti (Committee member)
Date: 2018-08
Keywords: Enhanced sparse coding
Feature enchancement
Image classification
Metric learning
Visual recognition
Availability: Unrestricted
Abstract:

Image classification is an active and rapidly expanding research area in computer vision and machine learning due to its broad applications. With the advent of big data, the need for robust image descriptors and learning methods to process a large number of images for different kinds of visual applications has greatly increased. Towards that end, this dissertation focuses on exploring new image descriptors and learning methods by incorporating important visual aspects and enhancing the feature representation in the discriminative space for advancing image classification.

First, an innovative sparse representation model using the complete marginal Fisher analysis (CMFA-SR) framework is proposed for improving the image classification performance. In particular, the complete marginal Fisher analysis method extracts the discriminatory features in both the column space of the local samples based within class scatter matrix and the null space of its transformed matrix. To further improve the classification capability, a discriminative sparse representation model is proposed by integrating a representation criterion such as the sparse representation and a discriminative criterion. Second, the discriminative dictionary distribution based sparse coding (DDSC) method is presented that utilizes both the discriminative and generative information to enhance the feature representation. Specifically, the dictionary distribution criterion reveals the class conditional probability of each dictionary item by using the dictionary distribution coefficients, and the discriminative criterion applies new within-class and between-class scatter matrices for discriminant analysis. Third, a fused color Fisher vector (FCFV) feature is developed by integrating the most expressive features of the DAISY Fisher vector (D-FV) feature, the WLD-SIFT Fisher vector (WS-FV) feature, and the SIFT-FV feature in different color spaces to capture the local, color, spatial, relative intensity, as well as the gradient orientation information. Furthermore, a sparse kernel manifold learner (SKML) method is applied to the FCFV features for learning a discriminative sparse representation by considering the local manifold structure and the label information based on the marginal Fisher criterion. Finally, a novel multiple anthropological Fisher kernel framework (M-AFK) is presented to extract and enhance the facial genetic features for kinship verification. The proposed method is derived by applying a novel similarity enhancement approach based on SIFT flow and learning an inheritable transformation on the multiple Fisher vector features that uses the criterion of minimizing the distance among the kinship samples and maximizing the distance among the non-kinship samples.

The effectiveness of the proposed methods is assessed on numerous image classification tasks, such as face recognition, kinship verification, scene classification, object classification, and computational fine art painting categorization. The experimental results on popular image datasets show the feasibility of the proposed methods.


If you have any questions please contact the ETD Team, libetd@njit.edu.

 
ETD Information
Digital Commons @ NJIT
Theses and DIssertations
ETD Policies & Procedures
ETD FAQ's
ETD home

Request a Scan
NDLTD

NJIT's ETD project was given an ACRL/NJ Technology Innovation Honorable Mention Award in spring 2003