Multi-layer recurrent neural network based offline Arabic handwriting recognition
Liren Chen, Ruijie Yan, Liangrui Peng, Akio Furuhata, Xiaoqing Ding · 2017
Offline Arabic handwriting recognition has been a challenging sequence modeling problem due to the cursive nature of Arabic script. This paper proposes a four-layer bidirectional Gated Recurrent Unit (GRU) network incorporated with dropout mechanism, which improves the model capacity and the generalization ability compared with a baseline system of a three-layer Long Short Term Memory (LSTM) network. Without using hand-crafted features and word matching, the proposed method finally outperforms other LSTM based segmentation-free methods on the IFN/ENIT database of hand-written Arabic words.