Printed New Tai Lue character recognition based on BP neural network

Abstract

With the development of optical character recognition research, many techniques for printed character recognition have been developed. But the character recognition study for the New Tai Lue has lagged behind. As a result, the digital processes of New Tai Lue have been in troubles. To solve this problem, this paper proposed a printed New Tai Lue character recognition method based on the Back Propagation (BP) Neural Network. Firstly, some image preprocessing including skew image correction, binarization are performed on the New Tai Lue images obtained by a scanner. Secondly, the character segmentation is carried out after the text line segmentation and the word segmentation processes. Thirdly, the feature vectors are built by using the local gray feature, the centroid coordinate and the original size of the image; Lastly, A BP neural network is trained with a training set contained 51740 samples. Then a test set contained 4917 samples is applied to test the performance of the proposed method. The experiment results show that 98.04% recognition rate has been obtained by the proposed method, which proved that this method is effective for the recognition of printed New Tai Lue characters.