Browse Prior Art Database

Pairwise Discrimination with Decomposition Matching in Online Handwriting Recognition

IP.com Disclosure Number: IPCOM000104204D
Original Publication Date: 1993-Mar-01
Included in the Prior Art Database: 2005-Mar-18
Document File: 4 page(s) / 116K

Publishing Venue

IBM

Related People

Tappert, CC: AUTHOR

Abstract

For many alphabets, there are characters that are similar to other characters. For example, the characters T and J (written with a top bar) are similar except that the lower stroke of the J is curved, X and Y are similar except that the first stroke of Y is shorter and does not extend far beyond the second, T and + are similar except for the relative position of the two strokes of the characters. A stroke is the writing from pen down to pen up.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 45% of the total text.

Pairwise Discrimination with Decomposition Matching in Online Handwriting Recognition

      For many alphabets, there are characters that are similar to
other characters.  For example, the characters T and J (written with
a top bar) are similar except that the lower stroke of the J is
curved, X and Y are similar except that the first stroke of Y is
shorter and does not extend far beyond the second, T and + are
similar except for the relative position of the two strokes of the
characters.  A stroke is the writing from pen down to pen up.

      Many online, handwriting recognition systems use elastic curve
matching to match an unknown character against prototype (template)
characters [1, 3-5, 7].  One of the problems of such recognizers is
that all portions of the character are weighted equally without
regard to the location of the important differences or features.
This can lead to difficulty in discriminating between similar
characters.  The figure shows how an unknown character "J" could be
recognized as the character "T".  Here, J and T are written with two
strokes.  In this case, unfortunately, the horizontal stroke of the
unknown J has a slant similar to that of prototype T and unlike the
slant of the corresponding stroke of prototype J.  If the
dissimilarity between the horizontal stroke of the unknown J and
prototype J is greater than that between the vertical (and curved)
stroke of the unknown J and the vertical stroke of prototype T, then
the unknown J is more similar to prototype T and is therefore
recognized as a T.

       Figure.  How unknown "J" can be recognized as "T".

      Accurate pairwise discrimination focuses the discriminating
measurements on the portion of the characters that contain the
important differences.  Disclosed here is a new method for pairwise
discrimination.  This method is most efficiently combined with the
signal-processing, prototype-matching technique of decomposition
matching [6].  It focuses the discrimination on the appropriate
strokes and stroke  relationships that contain the important
differences for pairwise discrimination.

      In a previously-disclosed decomposition formula [2]  a
character matching distance is computed from stroke matching
distances and stroke relationship terms.  Specifically, the
x-coordinate matching  distance of an unknown and prototype character
of l strokes is

D sup p = Sum from k=1 to l D sub k sup p here +Sum from j=1 to l-1
Sum from k=j+1 to l < <n sub j> <n sub k> > [  x bar sub j sup u - x
bar sub k sup u - x bar sub j sup p  + x bar sub k sup p ]  sup 2
here
                   ---------------------------------------
                                    Sum from k=1 to l n sub k here
where D sub k is the matching distance of stroke k, unknown stroke k
is a sequence of n sub  k coordinates,  lbrace (<x sub 1> , <y sub
1>), %(<x sub 2> , <y sub 2>), ellipsis, %(<x sub <n s...