Rough Document Interpretation by Perceptual Organization
Proc. SDIUT '2003 (Symposium on Document Image Understanding Technology)
Greenbelt, Maryland
Organized by the Institute for Advanced Computer Studies, U. Maryland.
Eric Saund, David Fleet, James Mahoney, Daniel Larner
Palo Alto Research Center
Abstract
Not all document images of interest are comprised of properly scanned, cleanly
printed text in a known language, formatted according to standard layout
conventions. We are often interested in the content of {\it rough documents}.
Rough documents include handwritten notes, sketches, drawings, annotations,
doodles, specialized notations, unconventional layouts, and poorly imaged
markings. Despite their unruly diversity, rough documents embody visual
structure which is accessible and exploitable by humans by virtue of our
powerful perceptual apparatus. We believe that to achieve breadth and depth in
interpreting images of rough documents, computer systems will require a
foundation of image analysis approximating the human visual stage postulated as
Perceptual Organization. This whitepaper motivates and outlines a research
program along these lines. A viewpoint that raises the identification of
visual perceptual structure as a primary objective helps to open a broader
range of tasks for document image analysis, beyond character-to-text
transcription.
Paper
Back to Eric Saund's Papers Page.
Back to Eric Saund's Home Page.