New MDLSTM-based designs with data augmentation for offline Arabic handwriting recognition

Although deep learning techniques have achieved promising results in several fields including healthcare, monitoring, and smart cities, their application in handwriting recognition shows limited results, especially for the offline Arabic handwritten script. Therefore, there is a need to enhance existing deep learning architectures. Moreover, the Multi-Dimensional Long Short-Term Memory (MDLSTM) leverages the LSTM model by replacing the single recurring connection with as many connections as there are spatiotemporal dimensions in the data. These connections allow the network to create a flexible internal context representation that is robust to local distortions. In this context, MDLSTM based architecture has been explored and three new architectures with Dropout, ReLU, and Maxout are proposed for offline Arabic handwriting recognition. Moreover, data augmentation approach has been applied to validate the proposed models. Indeed, a new dataset is developed by some morphological operations applied to the existing dataset “IFN/ENIT”. The experimental results show that our models outperform existing ones and the best accuracy of 92.59% was recorded with the MDLSTM-CTC-Maxout model trained with the original IFN/ENIT dataset. Moreover, data augmentation improves the MDLSTM-CTC-Maxout proposed model’s accuracy to reach 93.46%.