IITA 1996 Annual Report

Retrieval of Digital Images by Means of Content
http://maya.ctr.columbia.edu:8080

Objective

To provide fast, content-based search of large satellite image archives over the Internet.

Approach

Develop a system with the following features:

  • Satellite image server supports fast, content-based (feature) search of large image libraries.

  • Search speed is obtained through the use of highly compressed, low resolution images for initial culling, with selective search of full resolution data on only those images that survive the culling.

  • Searches are defined in a general-purpose query language. Users may invoke predefined queries, or write their own query scripts.

  • Support for off-the-shelf browsers that use web technology, as well as a custom browser.

  • The custom browser supports query specification, region and other parameter selection, and user-specified color assignments.
Accomplishments

Delivered a prototype sysem to the U.S. Forest Service. This system includes content-based queries for locating fires/fire scars.

Developed a progressive framework to efficiently represent, index and retrieve large amounts of satellite imagery.

Achieved a sppedup factor of more than 500 times for some content-based operations such as classification.

Developed recipes for a set of scenarios that require content-based queries such as global change detection, fire and fire scar locations.

Significance

This project seeks to demonstrate feasibility of efficient image indexing and retrieval for image databases larger than one petabyte.

Plans

Install Beta version at RSPAC; complete integration of image operators. Install instructional tools. This project expires in Fiscal Year 1998.

Point of Contact

Dr. Chung-Sheng Li
Image Information System
IBM T.J. Watson Research Center
P.O. Box 704
Yorktown Heights, NY 10598
Email: csli@watson.ibm.com
Phone: 914-784-6661
Fax: 914-784-7455