Multi-Domain Deep Convolutional Neural Network for Ancient Urdu Text Recognition System

Deep learning has achieved magnificent success in the field of pattern recognition. In recent years Urdu character recognition system significantly benefited from effectiveness deep convolutional neural network. Majority research on text are concentrated formal handwritten and printed document. this paper, we experimented Challenging issue ancient literature documents. Due to its cursiveness, complex word formation (ligatures), context-sensitivity, inadequate benchmark dataset, document is very difficult process compared work, first, generated a dataset by extracting recurrent ligatures an fatawa book. Secondly, categorized augment generate batches augmented images that improvise training efficiency classification accuracy. Finally, proposed multi-domain Convolutional Neural Network which integrates spatial domain frequency CNN learn modular relations between features originating two different networks train The experimental results show network with achieves averaged accuracy 97.8% outperforms other models class. also for literature, well-known datasets not appropriate verified our prepared dataset.