Optimal Recognition of Characters Within Characters in Online Handwrit Ing Recognition
Original Publication Date: 1991-Jan-01
Included in the Prior Art Database: 2005-Apr-01
For many alphabets, there are combinations of characters that are similar to other characters. For example, the numbers 1 and 3 written closely together look like B. Similar confusions exist, for example, between d and cl, H and 1-1. (Image Omitted)
Optimal Recognition of Characters Within Characters in
alphabets, there are combinations of characters
that are similar to other characters. For example, the numbers 1 and
3 written closely together look like B. Similar confusions exist, for
example, between d and cl, H and 1-1.
of a recent article (1) permits a new view of this
character-within-character problem. In that article, 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 where
Dk is the matching distance of stroke k, unknown stroke k is a
sequence of nk coordinates, , subscripts refer to strokes,
superscripts u and p to unknown and prototype, n to the number of
matched points of a stroke, and x is the x-coordinate of the center
of gravity. All computations here are presented only for x
coordinates; they are similar for y coordinates.
example of matching two unknown strokes against
prototypes for B and for 1 3. On the one hand, from formula 1 the
matching distance of a two-stroke unknown against a two-stroke
prototype of B is where the third term compares the relative
positions of the two strokes. On the other hand, the sum of the
matching distances of the unknown strokes against prototypes for 1
and 3 is . Now, if the prototypes for 1 and 3 are
identical to those for the strokes of B, then , and, most
importantly, . Since the best match is the smallest distance, a
recognition system that simply adds character matching distances
prefers 1 3 to B, and, in general, several small characters to fewer
large similarly shaped ones. This preference is characteristic of
some recognition systems.
use ad hoc methods to correct this problem.
example, the sum of the character matching distances has been
multiplied by the number of characters in the sequence to offset the
bias toward recognition of a greater number of characters .
formula, however, suggests a theoretically based
solution. Consider the problem of recognizing a sequence of
characters using formula 1. The matching distance for any unknown
sequence of characters against a prototype sequence of characters is
the sum of the stroke dista...