Research Of Handwritten Character Recognition Based On Neural Network

This paper studies handwritten numeral recognition. It proposes new boundary feature extraction way of handwritten numeral recognition, which is simple and effective. It doesn't need to thin, saves time and improves the running speed. Feature vectors are trained and tested by the neural network of RBF, BP and PNN. The paper not only tests the feature extraction, but also analyzes the results by different neural networks. By comparison, a conclusion is drawn that PNN is more suitable to train and test for pattern identification, which provides a powerful reference for actual use.New feature extraction in this paper is of wide applicability, which gives proper parameters according to different character set. Without affecting general recognition rate, one for certain type of character set is improved. In other words, the new way is more fit for concrete problem.