University of Illinois Urbana-Champaign

Cross-Categorization Transfer Learning Enhancing Image and Video Classification Performance

Chang, Shiyu

This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.

Permalink

https://hdl.handle.net/2142/46541

Description

Title
Cross-Categorization Transfer Learning Enhancing Image and Video Classification Performance
Author(s)
Chang, Shiyu
Contributor(s)
Huang, Thomas S.
Issue Date
2011-05
Keyword(s)
  • machine learning
  • image classification
  • video classification
  • cross-category transfer learning
Abstract
In this paper, we concentrate on exploring the cross-category knowledge to enhance the information on the target categories with a small number of positive training examples. In many cases, even the intra-category knowledge may still be insufficient due to the scarce positive samples of the target category. On the other hand, transferring the cross-category knowledge is appealing as a way to solve or alleviate this problem by exploring knowledge in correlated categories. It is a quite challenging problem due to the nature of semantic differences among categories. To approach such cross-category transfer learning (CCTL), we propose to explore the intrinsic correlations between the source and target categories. A cross-category label propagation process is developed to transfer the category information from the source to the target categories. Moreover, the proposed CCTL can automatically detect when to transfer, which plays a role of “safety valve” to avoid the transfer of cross-category knowledge that is harmful for modeling the target category. The experiments in real-world image and video data sets demonstrate the competitive results and illustrate how the transfer processes connect between the source and target categories.
Type of Resource
text
Language
en
Permalink
http://hdl.handle.net/2142/46541

Owning Collections

Senior Theses - Electrical and Computer Engineering PRIMARY
The best of ECE undergraduate research