Research on the Recognition of Offline Handwritten New Tai Lue Characters Based on Bidirectional LSTM

摘要

Deep learning has made breakthrough progress in image recognition, target detection and tracking in recent years. It is proved too good at classification tasks. In this paper, we have compared use of Convolutional neural network (CNN), VGG16, Long Short-Term Memory (LSTM), and Bidirectional LSTM to perform offline handwriting New Tai Lue Characters recognition. These methods have been tested on a dataset build by our laboratory. For testing purpose 58795 samples including 9834 test samples of handwriting New Tai Lue Characters are used in these experiments The experimental results show that the recognition rates are 91.23%, 89.33%, 92.78% for CNN, VGG16 and LSTM. Moreover, the best recognition result is obtained with the Bidirectional LSTM based method, whose recognition rate is 94.87% on the dataset.