Browse Prior Art Database

Noise Tolerance for Forms Recognition

IP.com Disclosure Number: IPCOM000111252D
Original Publication Date: 1994-Feb-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 4 page(s) / 228K

Publishing Venue

IBM

Related People

Milano, MF: AUTHOR

Abstract

The problem addressed by this invention involves recognizing that the scanned image of a form represents a particular type of form, and once this identification has been made, verifying that this identification is correct and calculating form skew (rotation of the scanned image of the input form) and form offset (shift of the input form image compared to the image used to define the form in a Forms Library). One way to perform these tasks is to analyze the line structure of the image and compare it to the line structures of the types of forms defined to the system in the Forms Library. An approach that relies on this method is documented within [*], hereafter referred to as IFP.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 24% of the total text.

Noise Tolerance for Forms Recognition

      The problem addressed by this invention involves recognizing
that the scanned image of a form represents a particular type of
form, and once this identification has been made, verifying that this
identification is correct and calculating form skew (rotation of the
scanned image of the input form) and form offset (shift of the input
form image compared to the image used to define the form in a Forms
Library).  One way to perform these tasks is to analyze the line
structure of the image and compare it to the line structures of the
types of forms defined to the system in the Forms Library.  An
approach that relies on this method is documented within [*],
hereafter referred to as IFP.  In IFP, the first step in the form
recognition/verification procedure is to process the entire form
image, creating a list of all the horizontal and vertical lines.  A
form is recognized by traversing a binary decision tree, created when
the set of recognizable form types is defined in the Forms Library,
comparing a subset of the lines in the image with those of the
defined form types.  Once the form has been recognized, it is
verified by comparing all of the horizontal and vertical lines in the
image with those of the selected form type.  The form skew is
calculated by choosing the median value from a list of the slopes of
all the lines on the input form, and the form offset is calculated by
choosing the median values from lists of the horizontal and vertical
offsets of all of the matching lines on the form.  These skew and
offset calculations are then used to adjust field coordinates during
the data extraction process in IFP.

      One limitation of IFP is that it was designed specifically to
handle specific type forms.  IFP is able to perform form recognition,
form verification, and form skew and offset calculations with a high
degree of accuracy when processing the limited set of forms using a
limited set of scanners.  However, if the set of forms and/or
scanners is broadened, IFP has much more difficulty performing these
tasks because it is not able to adequately handle the different
characteristics of the line structures of some forms and/or the
different types of noise or inconsistencies introduced by some
scanners.

      This invention describes several improvements to the line
structure analysis procedures within IFP which allow for more
accurate form recognition, form verification, and form skew and
offset calculations for a wider variety of forms using a wider
variety of scanners.

      There are three components to the solution to this problem,
each of which is designed to handle a particular type of difficulty
that is encountered with some forms and/or scanners.  Each component
improves upon a part of the process in IFP for finding the lines on
the input form and deciding which lines to use for form skew and
offset calculations.  Basically, IFP finds the lines on a form by
first...