| |
|
|
Electronic Systems Design Seminar
|
|
|
|
|
|
|
Despite a rich history of
research that uses sketch recognition to make drawings come to life, there are
few practical successes. One promising
approach is to disregard recognition during the drawing process, instead
emphasizing the value of "ink as ink" and presenting the user with
the abstraction of a blank sheet of paper.
Yet there are numerous ways that recognition can add value to the ink
after it has been captured, including but not limited to, better search,
archival, editing, and information sharing.
In this talk I motivate the
need for recognition systems that recognize ink strokes that have been laid
down on a paper-like surface, discerning structure after the fact so as not to
interrupt thought capture. I give three
examples of original domain-specific algorithms that obey this principle. The first is a recognizer which pulls
handwriting structure out of freeform notes and has been built into a shipping
commercial product. The second recognizes freeform annotation markup in a
structured document context. The third
is a more general ink parsing algorithm for context-free visual diagramming
languages.
Michael is a PhD candidate
in EECS at UC Berkeley. His work is in
the areas of sketch recognition algorithms, sketch user interfaces, and
computer aided design tools. As an intern
at Microsoft he was a key contributor to the first-generation ink parsing
algorithms of the Tablet PC and to the design of its parsing architecture.
