|
To clarify some of the jargon on our web pages, we present here some of
the terminology of handwriting recognition.
Writer-dependent Trained to
recognize the handwriting of a particular writer. cf. writer-independent
Writer-independent Recognizing
the handwriting of many different writers. cf. writer-dependent
Vocabulary, lexicon, word-list,
dictionary A list of words known to the recognizer. This can be
changed by the user for specific applications, or even for particular
fields in a data-entry tool. Typically we have 23,000 words in the
cursive note-taking application, 100,000 in the ThinkWrite product, and
fewer in a form-filling application.
Training To improve the accuracy on a
particular writer's handwriting, we collect some samples of handwriting
from the writer, and automatically adjust the parameters of the
recognition engine to arrive at a writer-dependent recognizer.
Adaptation Adaptation is training on
a small amount of data to quickly increase the accuracy when a new
writer starts to use the system. cf. Training
On-line On-line handwriting recognition
is recognition of handwriting written with an electronic stylus onto a
tablet or pen computer. From the digitizing device, the computer
receives a stream of coordinates describing the path of the stylus tip.
cf. Off-line.
Off-line Off-line handwriting
recognition is recognition of handwriting that has already been written
on a piece of paper, using a pen or pencil. The computer scans the paper
with an optical scanner and recognizes the handwriting from the image,
with no information about the timing or ordering of the strokes. cf.
On-line.
Discrete Discrete handwriting is where
the pen is lifted off the paper between each character. It is printing
as opposed to cursive writing. IBM's ThinkWrite handwriting
recognition product recognizes discrete handwriting, with occasional
run-on.
Cursive Cursive handwriting is `joined-up'
writing where consecutive letters are joined together. Recognition of
cursive writing is harder than discrete writing because it is very
difficult to work out where one letter ends and the next begins.
Unconstrained Unconstrained means
that no restriction is placed on the style of handwriting that the
recognizer will accept. In particular we use unonstrained to refer to a
recognizer that will recognize cursive and discrete handwriting.
Electronic ink The lines on the screen to
mark the path of an electronic pen. Words written in ink are on-line
handwriting and can be recognized and transcribed into text.
Hidden Markov model This is the technique
used by the cursive recognition engine. Statistics of training samples
are used by the model to recognize unseen words.
|
