Attention-Based CNN-RNN Arabic Text Recognition from Natural Scene Images

According to statistics, there are 422 million speakers of the Arabic language. Islam is second-largest religion in world, and its followers constitute approximately 25% world’s population. Since Holy Quran Arabic, nearly all Muslims understand language per some analytical information. Many countries have as their native official well. In recent years, number internet users speaking has been increased, but very little work on it due complications. It challenging build a robust recognition system (RS) for cursive nature languages such Arabic. These challenges become more complex if variations text size, fonts, colors, orientation, lighting conditions, noise within dataset, etc. To deal with them, deep learning models show noticeable results data modeling can handle large datasets. Convolutional neural networks (CNNs) recurrent (RNNs) select good features follow sequential technique. two offer impressive many research areas recognition, voice several tasks Natural Language Processing (NLP), others. This paper presents CNN-RNN model an attention mechanism image recognition. The takes input generates feature sequences through CNN. transferred bidirectional RNN obtain order. miss preprocessing segmentation. Therefore, used generate output, enabling relevant information from sequences. An implements end-to-end training standard backpropagation algorithm.