Workshop Summary
~~~~~~~~~~~~~~~~

1. Samples for alphabetic character training and testing, larger than
   "NIST Special Database 1", are needed.

2. OCR results which do not provide a continuous curve of reject percent
   versus character decision accuracy are of limited value. NIST has a
   draft set of standard definitions of these and other system testing
   errors available upon request.

3. Existing recognition accuracies are at commercially useful levels but
   rates desired by customers of 200 to 600 characters per second have not
   been achieved.

4. Each specific OCR systems application will required a test database
   tailored to that application to achieve good engineering design.