Image webs : discovering and using object-manifold structure in large-scale image collections

Placeholder Show Content


This dissertation explores a method for data-mining in large image collections with applications to object recognition and dataset visualization. The method builds graphs that preserve the local metric structure of the ideal object-manifolds by linking instances of the same object in different scenes using local image feature matching techniques. A range of image matching techniques are evaluated resulting in a design optimized for the task of preserving local metric structure. Benchmark evaluations on a large dataset show the proposed graph representation is useful for fine-grained semi-supervised object discovery, localization, and annotation tasks. These findings are interesting because fine-grained object recognition for a large number of objects is beyond the reach of the current state-of-the art computer vision algorithms. This work suggests promising directions for data-driven computer vision techniques for automatically organizing massive image collections in which the size of the dataset becomes an advantage rather than a limiting factor.


Type of resource text
Form electronic; electronic resource; remote
Extent 1 online resource.
Publication date 2013
Issuance monographic
Language English


Associated with Heath, Kyle
Associated with Stanford University, Department of Electrical Engineering.
Primary advisor Guibas, Leonidas J
Thesis advisor Guibas, Leonidas J
Thesis advisor Girod, Bernd
Thesis advisor Li, Fei Fei, 1976-
Advisor Girod, Bernd
Advisor Li, Fei Fei, 1976-


Genre Theses

Bibliographic information

Statement of responsibility Kyle Heath.
Note Submitted to the Department of Electrical Engineering.
Thesis Thesis (Ph.D.)--Stanford University, 2013.
Location electronic resource

Access conditions

© 2013 by Kyle Howard Heath
This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).

Also listed in

Loading usage metrics...