On-Line Hand-Written Chinese Character Recognition by an Improved Perceptron Considering the Order of Classification.
Takahumi Oohori, Tadanori Domi, Yu Rin, Kazuhisa Watanabe · IEEJ Transactions on Industry Applications · 1991
This paper proposes a new sequential perceptron, which is an improved version of the two-layered simple perceptron, and applies it to the on-line hand-written chinese character recognition problem.The basic idea of the sequential perceptron is as follows: The simple perceptron cannot classify the linearly non-separable pattern set which consists of several categories, but if the already classified categories are removed during learning process, some categories may become linearly separable and can be classified from the other categories.The sequential perceptron, during learning process, discriminates a locally and linearly separable category and removes it sequentially. During recognition process, it recognizes an unknown character by sequentially searching already learned and ordered characters.The proposed method can significantly improve the rate of learning and the rate of recognition for unknown data. It is verified by eighty kinds of on-line hand-written chinese characters recognition experiments.