Document Image Representations for Automatic Thumbnail Creation
Dr. Kathrin Berkner
Ricoh Research Center
Menlo Park, CA
Tuesday, February 25th, 12:30 PM, ENS 637
berkner@congo.crc.ricoh.com
Abstract
More and more paper documents are stored in electronic form. In general, that
includes scanning of a document and archiving it in some file system or
database. With a steadily increasing amount of digitally available documents the
problem of retrieval becomes important. Besides searching by meta data such
as creation time, file location, etc., small low-resolution representations
of the original images, so-called thumbnails, are often displayed in order
to give the user some visual help to find a desired document. A big problem
with current thumbnails is that the image content become unrecognizable, e.g.
text becomes unreadable.
In this talk a representation of a document is presented that uses a
combination of JPEG 2000-based image analysis and OCR-based text analysis to
derive resolution-sensitive representations of document. Using those
representations it is possible to create thumbnails that display
recognizable content for every thumbnail size.
Biography
Kathrin Berkner received her PhD in mathematics from the University of Bremen,
Germany, in 1996 and joined Ricoh Innovations in 1998 after being a
postdoctorial researcher at Rice University. She is a Senior Research
Scientist in the Color Image Processing group at the California Research
Center of Ricoh Innovations.
Her interests cover theory and applications of various types of wavelet
transforms, use of the JPEG2000 standard for applications beyond compression,
and new image representations.
A list of Wireless Networking and Communications Seminars is available at
from the ECE department Web pages under "Seminars".
The Web address for the Wireless Networking and Communications Seminars is
http://signal.ece.utexas.edu/seminars