| Summary: | 碩士 === 國立臺灣科技大學 === 電子工程學系 === 84 === In this thesis, a method of character preclassification for
handwritten Chinese character recognition is proposed. Since
the number of Chinese characters is very large (at least 5401s
for daily use), we employ two stages to reduce the candidates
of input character. In stage I, we try to extract the first
primitive features from handwritten Chinese characters and use
the fuzzy rules to create the four preclassification groups.
The purpose in stage I is to reduce the candidates roughly. In
stage II, we extract the second primitive features from
handwritten Chinese characters and then use the Supervised
Extended ART (SEART) as the classifier to generate the
preclassification classes for each preclassification group that
we create in stage I. The SEART classifier has excellent
performance, fast, good generalization and exceptions handling
ability in complex problems. Since the number of characters in
each preclassification class is smaller than that in the whole
character set, the problem becomes simpler. In order to
evaluate the proposed preclassification system, we use the 605
Chinese character categories in the text books of elementary
school as our training and testing data. The database used is
HCCRBASE (provided by CCL, ITRI, Taiwan). We select the even
samples of samples 1－100 as the training set, and the odd
samples of them as the testing set. The preclassification rate
that characters of testing set can be distributed into correct
preclassification classes is 98.11%.
|
|---|
