We need to recognize the fields in a handwritten insurance application
form. Are there libraries available that we can pass parts of the form
to and get back the recognized text? Ideally, the library would be
accessible from Smalltalk, but Java or C++-based libraries would also
work. What kind of accuracy is possible from this kind of approach?

---

Request for Question Clarification by hummer-ga on 19 Jul 2006 13:00 PDT

Hi max4321,

How's this...

2) FormAgent
FreeStyle? Recognition Engine
FreeStyle employs CerebralNet, the most advanced recognition
technology available ...The document may contain individual letters
and words, or even sentences and paragraphs. CerebralNet technology
integrates, in real time, image, context, and linguistic validation
rules to ensure the highest degree of accuracy possible."
The FreeStyle recognition engine is seamlessly comprehensive. It
recognizes both handwritten and machine-printed words simultaneously
and automatically. It also recognizes OMR, barcodes, logos, and form
identities."
"Besides the usual built-in validation rules such as lexical and
geometrical context to enhance accuracy, FormAgent uses a real-time
higher-level context known as data objects to assist the recognition
process. For example, FormAgent cross-checks the street, city, state,
and zip code of an address from the beginning of, rather than after
the recognition process. In this way, the system dramatically cuts
down on mistakes. An otherwise unreadable address can be correctly
read in most cases. Other data objects include dates, phone numbers,
and e-mail addresses. FreeStyle also uses a large dictionary to limit
candidates to recognized words. The Trueword module is built into our
recognition engine from the start rather than as a post-processing
tool, as other vendors often do.
Recognition results are stored in an HTML text file. This file
contains such information as the multiple choices of each character
and word, their confidences, geometric locations, etc. This
information can be easily accessed by FormAgent's other components and
modules. For example, Verify will use it to improve the efficiency of
the editing process.
CereSoft Verify PageMode Editor Verify
FormAgent will invoke a carefully designed editor to correct any
remaining recognition errors. The editor has two editing modes. In the
Field Mode, individual fields will be presented to users one at a time
for manual correction. But if field alone cannot resolve the ambiguity
of a word, Page Mode can be accessed by double-clicking the field
image. In Page Mode, the editing field and the original image in the
page are highlighted together for easy identification. Alternative
choices for each character and field are displayed at all times along
with the field you are editing.
Other Features 
FormAgent is completely scalable. It can distribute work to many PC
stations for scanning, recognition, verification and data export. It
is also Internet enabled, so that remote processing is possible over
private or public networks."
http://www.ceresoft.com/form.html

FormAgent Demo
http://www.ceresoft.com/demo.htm

For additional product information contact:
sales@ceresoft.com
http://www.ceresoft.com/map.html

...or this?

2) DataCap Handwriting Recognition
ICR Recognition Software
http://www.datacap.com/primer/recog/icr/

Webinar: ICR 101
Solving the Mystery of Handprint Recognition
http://www.datacap.com/webinar/icr101/

Regards,
hummer

max4321-ga

For a commercial application I would advise a commercial solution.
We've used the products from ABBYY Software House many times with
great success:

http://www.abbyy.com/

For a bespoke application you'll profit the quickest from using their
(Software Development Kits) SDK:

http://www.abbyy.com/sdk/

It could be an expensive road (in terms of time and money) to get that
110% just-the-way-you-like-it solution when one of their current
offerings may well get you 99% of what you want and a better ROI. We'd
suggest Form Reader:

http://www.abbyy.com/formreader/

Does this answer your question?

affordablemagic-ga