The form reader (FR) is used during the recognition step, in order to process
forms of a specified class.
User interaction with
the system is limited to provide to FR the class of forms to be processed,
feeding the scanner with forms and, if requested, correct the output of the
character recognition engine.
After the operator enters the form class, the FR starts to process the
incoming forms with the following sequence of operations.
Scan the form and convert it to an electronic image.
This is quite an ordinary operation which was also massively
carried out for creating a document data base.
Register the incoming form.
A critical point during the acquisition by the scanner is the
presence of skew, that is a rotation of the incoming page
with respect to the scanning lines. This may be due to
frictions of the scanner loading mechanism.
Skews, as well as page translations, may significantly increase
the difficulty of locating the fields.
Locate information fields into the incoming form.
After having registered the incoming form, the system must determine
the location of the information fields.
Portions of the image that are expected to contain the information
fields are isolated for the next recognition step.
Extract the information.
Once the system has discovered an area where an information field
is supposed to range, the recognition phase takes place.
We assume that the information is simply represented by printed
words, that are recognized as sequences of simple characters.
We adopt an optical character recognition system that has been
fully developed by our research group by using connectionist
models.
