Visual object recognition is a challenging problem with a wide range of real-life applications. The difficulty of this problem is due to variation in shape and appearance among objects within the same category, as well as varying viewing conditions, such as viewpoint, scale, illumination, occlusion and articulation of multi-part deformable objects. In addition, beyond the visual spectrum, depth and range sensors suffer from noise that inhibits object recognition. Under visual object recognition lie three subproblems that are each challenging: category recognition, instance recognition and pose estimation. Impressive work has been done in the last decade on developing systems for generic object recognition. Previous research has covered many recognition-related issues, however, the problem of multi-view recognition remains among the most fundamental challenges in computer vision. In this dissertation we focus on discovering low-dimensional latent representations that enable efficient joint multi-view object recognition over multiple modalities. These discovered latent representations allow us to work in lower dimensional latent spaces that capture the factors needed for object recognition from multi-view images and over multiple modalities; from images to depthmaps and 3D point clouds. Each of the models we present in this dissertation explore a different representation space of latent factors. The first model builds multiple kernel induced spaces to fuse information between different modalities and performs object pose estimation in a regression framework. The second model performs manifold analysis to solve categorization and pose estimation simultaneously. It does this by factorizing the space of topological mappings between a unified conceptual manifold and feature spaces. We present two variations of this; an unsupervised learning model and a supervised learning model. The third approach analyzes the representational spaces of the layers of Convolutional Neural Networks and builds on the findings by proposing a network that jointly solves category and pose. The fourth approach explores solving pose-invariant categorization of multi-part objects by shape information, in the form of 3D point clouds. We build a representation that inherently encodes pose and allows objects to be represented by multiple levels of object-part decompositions for more robust object recognition. In each approach we support our hypotheses by extensive experimentation.
Subject (authority = RUETD)
Topic
Computer Science
RelatedItem (type = host)
TitleInfo
Title
Rutgers University Electronic Theses and Dissertations
Identifier (type = RULIB)
ETD
Identifier
ETD_6938
PhysicalDescription
Form (authority = gmd)
electronic resource
InternetMediaType
application/pdf
InternetMediaType
text/xml
Extent
1 online resource (xviii, 152 p. : ill.)
Note (type = degree)
Ph.D.
Note (type = bibliography)
Includes bibliographical references
Subject (authority = ETD-LCSH)
Topic
Computer vision
Subject (authority = ETD-LCSH)
Topic
Pattern recognition systems
Subject (authority = ETD-LCSH)
Topic
Image processing--Digital techniques
Note (type = statement of responsibility)
by Tarek El-Gaaly
RelatedItem (type = host)
TitleInfo
Title
Graduate School - New Brunswick Electronic Theses and Dissertations
Identifier (type = local)
rucore19991600001
Location
PhysicalLocation (authority = marcorg); (displayLabel = Rutgers, The State University of New Jersey)
Rutgers University. Graduate School - New Brunswick
AssociatedObject
Type
License
Name
Author Agreement License
Detail
I hereby grant to the Rutgers University Libraries and to my school the non-exclusive right to archive, reproduce and distribute my thesis or dissertation, in whole or in part, and/or my abstract, in whole or in part, in and from an electronic format, subject to the release date subsequently stipulated in this submittal form and approved by my school. I represent and stipulate that the thesis or dissertation and its abstract are my original work, that they do not infringe or violate any rights of others, and that I make these grants as the sole owner of the rights to my thesis or dissertation and its abstract. I represent that I have obtained written permissions, when necessary, from the owner(s) of each third party copyrighted matter to be included in my thesis or dissertation and will supply copies of such upon request by my school. I acknowledge that RU ETD and my school will not distribute my thesis or dissertation or its abstract if, in their reasonable judgment, they believe all such rights have not been secured. I acknowledge that I retain ownership rights to the copyright of my work. I also retain the right to use all or part of this thesis or dissertation in future works, such as articles or books.