Convolutional Neural Network and BLSTM for Offline Arabic Handwriting Recognition

Abstract

There have been an exciting advance in machine learning during the last decade. In fact, increasing computer processing power has supported the analytical capabilities of recognition systems. In this study, we focus on Offline Arabic handwritten recognition and for this task, we propose a new system based on the integration of two deep neural networks. First a Convolutional Neural Network (CNN) to automatically extract features from raw images, then the Bidirectional Long Short-Term Memory (BLSTM) followed by a Connectionist Temporal Classification layer (CTC) for sequence labelling. We validate this model on an extended IFN/ENIT database, created with data augmentation techniques. This hybrid architecture results in appealing performance. It outperforms both hand-crafted features-based approaches and models based on automatic features extraction. According to the experiments results, the recognition rate reaches 92.21%.