Browse Prior Art Database

On-line Character Recognition Method using Static Features

IP.com Disclosure Number: IPCOM000110534D
Original Publication Date: 1992-Dec-01
Included in the Prior Art Database: 2005-Mar-25
Document File: 5 page(s) / 129K

Publishing Venue

IBM

Related People

Katoh, S: AUTHOR

Abstract

Disclosed is a candidate reduction method for on-line character recognition, using information on the outer shape of characters that is commonly provided by optical character readers. Fig. 1 shows the system flow chart. The size of input data is normalized in the second step. Preliminary classification 1 is a first step for reducing the number of candidates, using the number of strokes. Preliminary classification 2 and 3 are further candidate reduction steps, using the local connection direction feature and the peripheral feature, respectively. Details of these static features are given below. A discrimination step is applied to the output of the previous steps, and output candidates with distance values. Candidates are then sorted according to their distance values, and the top candidate is output as the correct one.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

On-line Character Recognition Method using Static Features

       Disclosed is a candidate reduction method for on-line
character recognition, using information on the outer shape of
characters that is commonly provided by optical character readers.
Fig. 1 shows the system flow chart.  The size of input data is
normalized in the second step.  Preliminary classification 1 is a
first step for reducing the number of candidates, using the number of
strokes.  Preliminary classification 2 and 3 are further candidate
reduction steps, using the local connection direction feature and the
peripheral feature, respectively.  Details of these static features
are given below.  A discrimination step is applied to the output of
the previous steps, and output candidates with distance values.
Candidates are then sorted according to their distance values, and
the top candidate is output as the correct one.
Process of Preliminary Classification 2

      This step decomposes the polyline of a stroke into the four
local elementary connections shown in Fig. 2.  The distance between
template and input data is defined as follows:
      Dz  = S  fi wi  - gij wi
where fi is the feature value of element i (i=1-4) of the input data,
gij is the template value of element i in category j, and wi is a
weighting factor.

      The following is an example showing how the four elementary
vector values are obtained.  After normalizing the observed data, we
can obtain time-series (x,y) data, such as (xo, yo), (x1, y1), (xz,
yz), Thus we can calculate the following dx, dy values:
      dxP  = xP+1 - xP
      dyP  = yP+1 - yP

      By applying the (dxP, dyP) vector to Fig. 3, we can obtain four
region numbers such as I - IV.  Then we can obtain the value to add
to the vector element by using the region number given in Table 1.
Finally, we calculate fi by summing all the increment values.

      By observing the relation between S fiwi and Dz of the correct
categories, we can define an experimental threshold function...