To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Unconstrained Arabic Handwritten Text Recognition Using Convolutional Recurrent Neural Network |
|
Author |
Ahmad AbdulQadir AlRababah, Mohammed Khalid Aljahdali |
| Citation |
Vol. 25 No. 1 pp. 135-140 |
|
Abstract |
Arabic text is cursive unlike many of most common languages, also, letters have different shapes depending on position of the letter, the shape of the first letter depends on what is after it, the shape of middle letters depend on what is before it and after it, and the shape of the last letter depends on what is before it. Moreover, different letters have very similar shapes. These properties make Arabic text recognition a difficult computer vision task. In this paper we try to solve the Arabic text recognition task without being constrained by a dictionary or a language model. We propose a new neural network architecture that achieves state of the art results in the unconstrained recogni-tion task. Our architecture is convolutional neural network with residual connections, followed by Bi-directional Long Short-Term Memory (BLSTM) layer, then finally a fully connected layer. |
|
Keywords |
deep learning; text recognition; convolutional neu-ral network; recurrent neural network; computer vision |
|
URL |
