Prototype Selection Methods for On-line HWR
Résumé
Prototype matching strategies such as DP-matching are powerful methods for character recognition problems with very large number of classes such as on-line Chinese character recognition. As for many problems with a large number of classes data is normally comparatively scarce for on-line Chinese characters and therefore prototype se- lection methods for use under these circumstances need to be robust against over-training. The methods investigated in this paper are incremental in the sense that all proto- type additions are based on the number of misclassifica- tions caused by the current database. This may be highly beneficial as it is possible to use the methods on databases that have undergone other optimization methods. Experi- ments have been conducted on the HANDS Kanji data and it reveals some interesting results.
Loading...