Arabic words Recognition using CNN and TNN on a Smartphone

Arabic script recognition has been a challenging due to the variability of writing styles, to the nature of Arabic scripts, to the complexities of processing steps and to the varieties of recognition methods. This paper uses a Convolutional Neural Network (CNN) for character recognition and Transparent Neural Network (TNN) for words reading. Because Arabic character segmentation is a very complicated step, we recognize only the first, the last character of all connected components of the recognized word and the isolated ones. A combination between the CNN and the TNN will complete the recognition of the whole word. CNN is a multi-layer feed-forward neural network that extracts features and properties from the input data. TNN is a special NN that recognize words from already activated characters and part of words. These methods are already used on computer recognition system. The proposed work is to integrate these methods and adapt them to the android operating system to apply them on smartphone. The evaluation is done on a database of Signboards Images of printed town names and the recognition rate is 98%.