|
|
|
The
Earth
and Space Science Browser (ESSB)
is the prototype system that we use to explore different image retrieval
techniques. The ESSB is a web- accessible database system
that we use to develop image retrieval methods that address the
needs of learners and novice users.
The database contains approximately 1400 Earth and Space
Science images, text information for those images, and content information derived from
those images. The text descriptions were compiled by UM School of Information
students and provide valuable background information about the images. We have implemented
several computer vision algorithms to compute the image content information. These
algorithms use raw image data to determine the visual characteristics of the images. For
example, the database contains image content information that describes the color, shape,
and texture characteristics of each image.
The ESSB user interface is constantly being improved
and enhanced. The current version of this interface supports several different search
modalities. For example, the interface features an image classification hierarchy, based
on the text information, that users can browse to find groups of thumbnail images or
specific images. The user interface also supports keyword search. In addition to these
features, we have implemented several similarity search engines for the user interface.
These engines allow the user to define a set of query images. Given a set of query images,
the similarity engines return images whose visual attributes are similar to the query
images.
We continue to explore new ways to utilize the
different types of information in the ESSB database. Ideally, the combination of image
content information and text descriptions will enable us to develop better ways to search
for images. Our plans for future work include: implementing a relevance feedback system,
using both text descriptions and image content information for search, and implementing
methods of content based retrieval that do not require a seed set images. |
