PENGENALAN TULISAN TANGAN KARAKTER HIRAGANA MENGGUNAKAN DCT, DWT, DAN K-NEAREST NEIGHBOR

Abstract

Research to recognize hiraga
na character based image processing has been widely practiced and
even the accuracy level is close to 100%. However, the input image that used is still in the form of
japanese characters print
-
out while the handwriting has not been studied. So in this stu
dy tested the
recognition of hiragana letters derived from handwriting format. Jpeg. Of the several related
studies, the most commonly used compression approach for JPEG images is the DCT and DWT
algorithms, so both algorithms are used in this study to be
tested and compared their performance.
In the system tested 45 images of 3 people handwriting hiragana character with KNN
-
based
classification where previously 45 different images of the 3 people are trained by each DWT and
DCT algorithms. The result is ba
sed on the distance parameters that exist in the KNN algorithm,
the DWT algorithm is superior to the DCT algorithm. The achievement of the maximum accuracy
level obtained for each DWT
-
DCT algorithm is on the cityblock distance parameter 82.61%
(DWT) and co
rrelation distance 58.70% (DCT).