Image Descriptions for Browsing and Retrieval
Abstract
The main thrust of our work under this grant has been the definition of basic image representations that are most appropriate for image retrieval. With the aim of a unified treatment, we have developed the notion of a signature to represent the color, shape, or texture content of an image. We have proposed a new, perceptually motivated metric for signatures, the Earth-Mover's Distance (EMO), and designed sound and efficient algorithms for its computation. Based on these foundational elements we have developed the notion of database navigation as a novel, effective paradigm for interaction with a large database of images. These concepts were the basis for a solid list of publications. We have demonstrated all our ideas with a series of software systems for the retrieval of images from large repertories based on color, texture, shape, and facial features. Eight PhD students have been funded by this project, and three of them have completed their PhD degrees at Stanford with theses on various aspects of image retrieval.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2000
- Accession Number
- ADA379138
Entities
People
- Carlo Tomasi
- Leonidas J. Guibas
Organizations
- Stanford University