|
In this thesis, a new coarse classification scheme which
makes use of structure feature for on-line Chinese character
recognition is presented. Six structure types are defined in
this thesis. The main purpose of coarse classification is to
classify on-line Chinese characters into one of the six
structure types and divide them into separated parts
accordingly, if they can be divided. Our proposed coarse
classification scheme can be used to classify the following
three writing style categories : square writing characters,
partial connected characters, and script characters. Two coarse
classification methods are proposed in this thesis. One is
the line segment order method, and another is the projection
method. The line segment order method makes use of the line
segment order embedded in the input character to judge which
kind of structure it belongs to. The projection method makes
use of the projection of the line segments of the considered
character onto the x and y axis to judge which kind of
structure it belongs to. The line segment order method can
classify most of the characters presented in our coarse
classification process, and the projection method is utilized
as the auxiliary method to resolve the ambiguity which can not
be classified using the line segment order method. Fourteen
databases of 5400 frequently used Chinese characters are
tested in our experiment. They include five sets of
square writing characters, two sets of partial connected
characters, and seven sets of full connected
characters. The average recognition rates of square writing
characters, partial connected characters, and full connected
characters are 96.7%, 92.8%, and 91.7%, respectively. The
experimental results show that our coarse classification
scheme is very effective and promising.
|