Abstract:
Urdu Nasteleeq text recognition is one of the very challenging problems in document image processing. The cursive nature of Urdu script makes character segmentation very difficult. Therefore, most of the researchers have shifted the focus on segmentation free approaches based on Urdu ligatures. In most cases, these ligatures are characterized using complicated and extensive feature extraction techniques. These features might fail to capture the minor details and hence lead to the loss of useful information. This study proposes the use of Convolutional Neural Networks for recognition of Urdu ligatures. Such deep learning techniques are novel and fast as compared to the conventional feature extraction methods. The input to the system are fixed size ligature images. The system automatically extracts features from raw pixel values of these images. The system evaluated on 18,000 Urdu ligatures with 98 different classes realized a recognition rate of up to 95%.
